CN104700099A - Method and device for recognizing traffic signs - Google Patents
Method and device for recognizing traffic signs Download PDFInfo
- Publication number
- CN104700099A CN104700099A CN201510150525.8A CN201510150525A CN104700099A CN 104700099 A CN104700099 A CN 104700099A CN 201510150525 A CN201510150525 A CN 201510150525A CN 104700099 A CN104700099 A CN 104700099A
- Authority
- CN
- China
- Prior art keywords
- traffic sign
- image
- training
- scanning window
- eigenwert
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/582—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of traffic signs
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a method and device for recognizing traffic signs. According to a detailed implementation mode, the method includes the steps that feature values, obtained through a preset feature algorithm, of scanned window images obtained through panorama spherical image division at a preset integral channel are obtained; according to the feature values of the scanned window images and a pre-trained detection classifier model, the scanned window images are detected to obtain traffic sign window images to be confirmed, wherein the pre-trained detection classifier model is obtained according to samples of the scanned window images and the feature values of the scanned window images through training; according to a pre-trained convolution neural network mode, the traffic sign window images to be confirmed are recognized, so that the categories of the traffic signs are obtained, wherein the pre-trained convolution neural network model is obtained through training according to samples of the determined traffic sign window images and the categories of the traffic signs. The accuracy of detecting and recognizing the traffic signs in panorama images is improved, and meanwhile the efficiency of updating road network data is improved.
Description
Technical field
The application relates to field of computer technology, is specifically related to computer object recognition technology field, particularly relates to the method and apparatus identifying traffic sign.
Background technology
The comprehensive navigation data in order to provide accurate, needs to identify the traffic sign in traffic network.
The method of traditional identification traffic sign, main dependence artificial treatment, in advance specific program software and base map data portion are deployed to and gather on car, by the outdoor workers through training with car operation: after observing traffic sign label, the distance of type and range estimation is manually entered in software by operator; After field data acquisition terminates, then by the base map before and after surveyor comparisons, by effective information updating to Traffic network database.The support of this process need various software instrument, but the accuracy of outcome data depends primarily on profile and the focus of operating personnel, and meanwhile, complicated collecting flowchart reduces the renewal efficiency of road net data.
Summary of the invention
The object of the application is to propose a kind of method and apparatus identifying traffic sign, solves the technical matters that above background technology part is mentioned.
On the one hand, this application provides a kind of method identifying traffic sign, described method comprises: obtain and divide by panorama spherical diagram picture the eigenwert that the scanning window image that obtains obtained by predetermined characteristic algorithm at predetermined integral passage; According to the eigenwert of described scanning window image and the detection sorter model of training in advance, detect scanning window image, obtain traffic sign video in window to be confirmed, wherein, the detection sorter model of described training in advance obtains according to the sample of scanning window image and eigenwert training thereof; According to the convolutional neural networks model of training in advance, identify traffic sign video in window to be confirmed, obtain traffic sign classification, wherein, the convolutional neural networks model of described training in advance obtains according to the sample of confirmed traffic sign video in window and the training of traffic sign classification thereof.
Second aspect, this application provides a kind of device identifying traffic sign, described device comprises: characteristic value acquisition module, divides the eigenwert that the scanning window image that obtains obtained by predetermined characteristic algorithm at predetermined integral passage for obtaining by panorama spherical diagram picture; Road traffic sign detection module, for according to the eigenwert of described scanning window image and the detection sorter model of training in advance, detect scanning window image, obtain traffic sign video in window to be confirmed, wherein, the detection sorter model of described training in advance obtains according to the sample of scanning window image and eigenwert training thereof; Traffic Sign Recognition module, for the convolutional neural networks model according to training in advance, identify traffic sign video in window to be confirmed, obtain traffic sign classification, wherein, the convolutional neural networks model of described training in advance obtains according to the sample of confirmed traffic sign video in window and the training of traffic sign classification thereof.
The method and apparatus of the identification traffic sign that the application provides, divide by panorama spherical diagram picture the eigenwert that the scanning window image that obtains obtained by predetermined characteristic algorithm at predetermined integral passage by obtaining, subsequently according to the eigenwert of described scanning window image and the detection sorter model of training in advance, detect scanning window image, obtain traffic sign video in window to be confirmed, then according to the convolutional neural networks model of training in advance, identify traffic sign video in window to be confirmed, obtain traffic sign classification, achieve and detect traffic sign video in window to be confirmed by the detection sorter model of training in advance, traffic sign classification is gone out by the convolutional neural networks Model Identification of training in advance, improve the accuracy of detection and Identification traffic sign in panoramic picture, improve the renewal efficiency of road net data simultaneously.
Accompanying drawing explanation
By reading the detailed description done non-limiting example done with reference to the following drawings, the other features, objects and advantages of the application will become more obvious:
Fig. 1 shows the exemplary process diagram of the method for the identification traffic sign according to the embodiment of the present application;
Fig. 2 shows the exemplary process diagram of the method for the detection model of the training training in advance according to the embodiment of the present application;
Fig. 3 shows a kind of exemplary process diagram of the method according to the acquisition of the embodiment of the present application traffic sign to be confirmed video in window;
Fig. 4 shows the exemplary process diagram of the method for the convolutional neural networks model of the training training in advance according to the embodiment of the present application;
Fig. 5 shows a kind of exemplary block diagram of the convolutional neural networks model preset according to the embodiment of the present application;
Fig. 6 shows the topology example figure of the device of the identification traffic sign according to the embodiment of the present application;
Fig. 7 shows the exemplary block diagram of the detection sorter model trainer according to the embodiment of the present application;
Fig. 8 shows the exemplary block diagram of the device for obtaining traffic sign video in window to be confirmed;
Fig. 9 shows the exemplary block diagram of the device of the convolutional neural networks model for training training in advance according to the embodiment of the present application.
Embodiment
Below in conjunction with drawings and Examples, the application is described in further detail.Be understandable that, specific embodiment described herein is only for explaining related invention, but not the restriction to this invention.It also should be noted that, for convenience of description, in accompanying drawing, illustrate only the part relevant to Invention.
It should be noted that, when not conflicting, the embodiment in the application and the feature in embodiment can combine mutually.Below with reference to the accompanying drawings and describe the application in detail in conjunction with the embodiments.
Fig. 1 shows the exemplary process diagram of the method for the identification traffic sign according to the embodiment of the present application.
As shown in Figure 1, identify that the method 100 of traffic sign can comprise:
Step 110, obtains and divides by panorama spherical diagram picture the eigenwert that the scanning window image that obtains obtained by predetermined characteristic algorithm at predetermined integral passage.
Here, first the panorama spherical diagram picture needing to carry out Traffic Sign Recognition can be obtained, then by the window of pre-set dimension, panorama spherical diagram picture is scanned, thus obtain scanning window image, obtain scanning window image afterwards again in the eigenwert that obtain of predetermined integral passage by predetermined characteristic algorithm.
Wherein, panorama spherical diagram similarly is the image that can show panorama referring to be spliced by multiple fish eye images.The window of pre-set dimension is generally the window preset according to the size of target object.
It should be noted that, when obtaining scanning window image in the eigenwert that predetermined integrating channel is obtained by predetermined characteristics algorithm, predetermined integrating channel can be image processing field any one or more integrating channel of the prior art, any one or more integrating channel that also can be or may develop in WeiLai Technology; Predetermined characteristics algorithm can be any one or more characteristics algorithm in image processing field prior art, also can for any one or more characteristics algorithm that may develop in WeiLai Technology.The mode of the application to concrete acquisition eigenwert does not limit, and it can be selected according to actual user demand by user.Such as, Ha Er (haar) eigenwert can be obtained by Ha Er algorithm at gray level image passage, histogram feature value is obtained by histogram feature value-based algorithm at the gradient image passage of different angles parameter, simultaneously random to eigenwert etc. by obtaining eigenwert algorithm at random at red, green, blue monochrome image passage.
In order to improve the speed of the eigenwert obtaining scanning window image further, the eigenwert of scanning window image can be obtained: alternatively by integrogram, in the above-described embodiments, step 110 can comprise: step 111, obtains the integrogram being divided the scanning window image obtained by panorama spherical diagram picture at the integrogram of predetermined integral passage; And step 112, according to the integrogram of scanning window image, obtain the eigenwert of scanning window image.
The eigenwert of scanning window image is calculated by the integrogram of above-mentioned use panorama spherical diagram picture, can the eigenwert of speed-up computation scanning window image, improve the counting yield of the eigenwert of scanning window image.
Alternatively, in order to the comparison of the eigenwert of the sample of the traffic sign video in window in the eigenwert of accelerated scan video in window and the monitoring sorter model of training in advance, can according to traffic sign classification, the predetermined integral passage of the eigenwert determining the sample obtaining traffic sign video in window and the eigenwert obtaining scanning window image and predetermined characteristic algorithm.Such as ban class mark as speed limit etc., the description on red channel can be strengthened.Preferably, predetermined integral passage can comprise following one or more: gray level image passage, red, green, blue monochrome image passage, the gradient image passage of different angles parameter, and edge-detected image passage etc.; And predetermined characteristic algorithm can comprise following one or more: Lis Hartel levies algorithm, local binary patterns characteristics algorithm, histogram feature algorithm and random to characteristics algorithm etc.It will be appreciated by those skilled in the art that predetermined integral passage in above-described embodiment and predetermined characteristic algorithm can combine as required, with the eigenwert required for obtaining carrying out detecting scanning window image.
The eigenwert of the sample of the above-mentioned acquisition traffic sign video in window determined according to traffic sign classification and obtain the predetermined integral passage of eigenwert and the predetermined characteristic algorithm of scanning window image, the characteristic of target can be described from multiple different angle, thus overcome target because the change that brings of different angles and illumination.
Step 120, according to the eigenwert of scanning window image and the detection sorter model of training in advance, detect scanning window image, obtain traffic sign video in window to be confirmed, wherein, the detection sorter model of training in advance obtains according to the sample of scanning window image and eigenwert training thereof.
Here, the detection sorter model of training in advance, first receive artificial mark, determine comprise the scanning window image of traffic sign and do not comprise the sample of scanning window image as traffic sign video in window of traffic sign in scanning window image, obtain the eigenwert of sample afterwards, then use sample and eigenwert thereof, the parameter of the sorter model set according to the actual requirements is trained, thus the detection sorter model of the training in advance obtained.
After the eigenwert getting scanning window image in above-mentioned steps 101, just can according to the detection sorter model of the eigenwert of scanning window image and above-mentioned training in advance, scanning window image is detected, to obtain the video in window of wherein doubtful traffic sign as traffic sign video in window to be confirmed.
Step S130, according to the convolutional neural networks model of training in advance, identify traffic sign video in window to be confirmed, obtain traffic sign classification, wherein, the convolutional neural networks model of training in advance obtains according to the sample of confirmed traffic sign video in window and the training of traffic sign classification thereof.
Here, the convolutional neural networks model of training in advance, first receive artificial mark, the sample detecting the traffic sign video in window to be confirmed that sorter model detects is classified, be labeled as concrete traffic sign classification or non-traffic classification, use sample and the traffic sign classification thereof of traffic sign video in window to be confirmed afterwards, to the convolutional neural networks model training set according to actual needs, thus the convolutional neural networks model of the training in advance obtained.
After get traffic sign video in window to be confirmed in above-mentioned steps 102, just can according to the convolutional neural networks model of traffic sign video in window to be confirmed and above-mentioned training in advance, traffic sign video in window to be confirmed is identified, to obtain the traffic sign classification of maximum probability corresponding to traffic sign video in window to be confirmed as the traffic sign classification identified.
The method of the identification traffic sign of the above embodiments of the present application, improves the accuracy of detection and Identification traffic sign in panoramic picture, improves the renewal efficiency of road net data simultaneously.
The method of the detection model of training training in advance is described below in conjunction with Fig. 2.
Fig. 2 shows the exemplary process diagram of the method for the detection model of the training training in advance according to the embodiment of the present application.
As shown in Figure 2, the method 200 of the detection sorter model of training in advance is trained to comprise:
In step 201, obtain the positive sample in the sample of scanning window image and negative sample, wherein positive sample comprises the video in window of traffic sign or comprises the video in window of traffic sign and surrounding expansion presetted pixel thereof, and negative sample comprises the scanning window image removing positive sample.
Here, positive sample in the sample of scanning window image and negative sample, first in scanning window image, obtain sample, afterwards according to the artificial mark received, the video in window sample of above-mentioned scanning window image being comprised traffic sign or the video in window comprising traffic sign and surrounding expansion presetted pixel thereof, as positive sample, will remove the scanning window image of positive sample as negative sample in sample.
Above-mentioned positive sample, when only comprising the video in window of traffic sign, can improve the efficiency of the eigenwert calculating positive sample.Consider that target object surrounding pixel also can effectively describe target object itself, therefore positive sample is when comprising the video in window of traffic sign and around expansion presetted pixel thereof, can improve the accuracy of the eigenwert describing positive sample.
In step 202., the eigenwert that positive sample and negative sample are obtained by predetermined characteristic algorithm at predetermined integral passage is obtained.
Here, when the positive sample of acquisition and negative sample are in the eigenwert that predetermined integral passage is obtained by predetermined characteristic algorithm, predetermined integrating channel can be image processing field any one or more integrating channel of the prior art, any one or more integrating channel that also can be or may develop in WeiLai Technology; Predetermined characteristics algorithm can be any one or more characteristics algorithm in image processing field prior art, also can for any one or more characteristics algorithm that may develop in WeiLai Technology.The application does not limit the predetermined integral passage of concrete acquisition eigenwert and predetermined characteristic algorithm, and it can be selected according to actual user demand by user.Such as, Ha Er (haar) eigenwert of algorithm acquisition is levied by Lis Hartel at gray level image passage, in the histogram feature value that the gradient image channel histogram characteristics algorithm of different angles parameter obtains, red, green, blue monochrome image passage by random to characteristics algorithm obtain random to eigenwert etc.
Alternatively, when the concrete calculating of eigenwert carrying out sample, in order to fast and effectively obtain the eigenwert that positive sample and negative sample obtained by predetermined characteristic algorithm at predetermined integral passage, according to traffic sign classification, predetermined integral passage and the predetermined characteristic algorithm of the sample of traffic sign video in window can be determined.Such as ban class mark as speed limit etc., the description on red channel can be strengthened.Preferably, predetermined integral passage can comprise following one or more: gray level image passage, red, green, blue monochrome image passage, the gradient image passage of different angles parameter, and edge-detected image passage; And predetermined characteristic algorithm can comprise following one or more: Lis Hartel levies algorithm, local binary patterns characteristics algorithm, histogram feature algorithm and random to characteristics algorithm.It will be appreciated by those skilled in the art that predetermined integral passage in above-described embodiment and predetermined characteristic algorithm can combine as required, to obtain the eigenwert that positive sample and negative sample are obtained by predetermined characteristic algorithm at predetermined integral passage.
In step 203, according to positive sample and negative sample and the eigenwert that obtained by predetermined characteristic algorithm at predetermined integral passage thereof, detect sorter model by boosting Algorithm for Training, obtain the detection sorter model of training in advance.
Here, positive sample and negative sample is obtained after the eigenwert that predetermined integral passage is obtained by predetermined characteristic algorithm in above-mentioned steps 202, sorter model can be detected by boosting Algorithm for Training, thus obtain the detection sorter model of result sorter as training in advance of a pin-point accuracy.
Below in conjunction with Fig. 3, the basis being described in above-described embodiment obtains the method for traffic sign video in window to be confirmed.
Fig. 3 shows a kind of exemplary process diagram of the method according to the acquisition of the embodiment of the present application traffic sign to be confirmed video in window.
As shown in Figure 3, the method 300 obtaining traffic sign video in window to be confirmed comprises:
Step 301, down-sampled continuously to panorama spherical diagram picture, obtain image pyramid.
Here, pyramid transform can be carried out to panorama spherical diagram picture, such as Laplacian Pyramid Transform etc., panorama spherical diagram picture be transformed to different metric spaces from life size, thus obtains the image pyramid comprising multi-level images.
Step 302, obtains the integrogram being divided the scanning window image of every grade of image in the image pyramid obtained by image pyramid at the integrogram of predetermined integral passage.
After above-mentioned steps 301 obtains image pyramid, first integration can be carried out to every first order image of image pyramid at predetermined integral passage, obtain the integrogram of image pyramid, divided by the integrogram of window to image pyramid of pre-set dimension afterwards, obtain the integrogram of the scanning window image of every grade of image.
Step 303, according to the integrogram of the scanning window image of every grade of image, obtains the eigenwert of the scanning window image of every grade of image.
After above-mentioned steps 302 obtains the integrogram of scanning window image of every grade of image, can according to the integrogram of the scanning window image of every grade of image, the integrogram of the scanning window image of every grade of image is utilized to carry out computing, the eigenwert of the scanning window image of quick obtaining every grade image.
Step 304, according to the eigenwert of the scanning window image of every grade of image and the detection sorter model of training in advance, detects the scanning window image of every grade of image, obtains traffic sign video in window to be confirmed.
After above-mentioned steps 303 obtains the eigenwert of scanning window image of every grade of image, can according to the eigenwert of the scanning window image of every grade of image, the scanning window image of every grade of image is detected by the detection sorter model of training in advance, obtain doubtful traffic sign video in window wherein as traffic sign video in window to be confirmed, thus obtain institute's subject to confirmation traffic sign video in window in image pyramid.
The method of the acquisition traffic sign video in window to be confirmed of the above embodiments of the present application, scanning window image is obtained in the pyramid diagram picture that panorama spherical diagram picture is obtained by pyramid transform, detect the scanning window image of every grade of image afterwards to obtain traffic sign video in window to be confirmed, decrease the possibility of omitting traffic sign window to be confirmed in panorama spherical diagram picture, therefore improve the accuracy obtaining traffic sign window to be confirmed.Obtain again the eigenwert of the scanning window image of every grade of image owing to have employed integrogram, improve the acquisition speed of eigenwert.
Below in conjunction with Fig. 4, the method for the convolutional neural networks model of training training in advance is described.
Fig. 4 shows the exemplary process diagram of the method for the convolutional neural networks model of the training training in advance according to the embodiment of the present application.
As shown in Figure 4, the method 400 of the convolutional neural networks model of training in advance is trained to comprise:
Step 401, according to Gaussian distribution, the convolutional layer of the convolutional neural networks model that initialization is preset and the weight of full articulamentum, wherein, the convolutional neural networks model preset comprises the convolutional layer, abstraction, layer, full articulamentum and the normalization layer that connect successively.
Here, the convolutional layer connected successively, abstraction, layer, full articulamentum and normalization layer can comprise: a convolutional layer, abstraction, layer, a full articulamentum and a normalization layer; Also can comprise multiple convolutional layer and with convolutional layer abstraction, layer, more than one full articulamentum and a normalization layer one to one.
The convolutional layer wherein arranged, can pass through convolution algorithm, original signal feature be strengthened, and reduces noise; The abstraction, layer arranged, can utilize the principle of image local correlation, carries out son sampling to image, while minimizing data processing amount, retain useful information.
Below for Fig. 5, the convolutional neural networks model preset is described.
Fig. 5 shows a kind of exemplary block diagram of the convolutional neural networks model preset according to the embodiment of the present application.
As shown in Figure 5, the convolutional neural networks model 500 preset comprises: the convolutional layer conv1, the abstraction, layer pool1 that connect successively, convolutional layer conv2, abstraction, layer pool2, convolutional layer conv3, abstraction, layer pool3, full layer fc1 in succession, full articulamentum fc2.
Wherein, conv1 has 16 sizes to be the convolution kernel of 5*5*3, and conv2 has 32 sizes to be the convolution kernel of 5*5*16, and conv3 has 64 sizes to be the convolution kernel of 5*5*32.Two full articulamentums respectively have 512 and 120 neurons.
Alternatively, in order to increase the output difference of similar purpose and non-similar purpose, default convolutional neural networks model can also comprise loss function layer.Such as, loss function can calculate with multiclass logarithm loss logarithmic loss function, namely
Wherein L is loss function, and N is training sample quantity, and M is categorical measure, p
i, ji-th sample for network output is the probability of jth class, y
i, jfor sample true value, if i-th sample belongs to jth class, be then 1, otherwise be 0.
Return Fig. 4, step 402, according to sample and the traffic sign classification thereof of confirmed traffic sign video in window, carry out iteration by the weight of error back propagation BP algorithm to convolutional layer and full articulamentum.
Wherein, the sample of confirmed traffic sign video in window is in the sample extracted in traffic sign video in window to be confirmed, the sample comprising the video in window of traffic sign after confirming according to the artificial mark received.
Further, the sample of confirmed traffic sign video in window can comprise: the original sample having confirmed traffic sign video in window, and has confirmed that the sample of traffic sign video in window to carry out after following one or more process normalizing to the image of preset window size by original: rotate, Pan and Zoom.By exptended sample, the accuracy rate of specimen discerning can be improved.
Corresponding with arranging loss function layer in step 401, carrying out iteration by the weight of error back propagation BP algorithm to convolutional layer and full articulamentum can comprise: by loss function and BP algorithm, the weight to convolutional layer and full articulamentum carries out iteration respectively.Such as, after all weights of above-mentioned Gaussian distribution initialization, stochastic gradient descent algorithm is adopted to carry out successive ignition to network model, every iteration, once first by the loss function L of above-mentioned loss function formula (a) forward computational grid, then oppositely successively calculates L relative to every layer of weights W
igradient, afterwards according to every layer of weights W
igradient updating weights W
i, namely
Wherein, α is the learning rate of default Gradient Descent,
for loss function L is relative to weights W
igradient.
Afterwards, the convolutional neural networks model of optimal weights can be determined by step S403 or step S404.
Step S403: if the difference of the weight after the weight after current iteration and last iteration is less than preset value, then the weight after current iteration is defined as optimal weights.
Step S404: if the weight after there is the iteration that error rate is minimum, then the weight after iteration minimum for error rate is defined as optimal weights.
Step S405: the convolutional neural networks model by the convolutional neural networks model specification comprising optimal weights being training in advance.
After the convolutional neural networks model training training in advance, then the step 103 in Fig. 1 can comprise: by traffic sign video in window input convolutional neural networks model to be confirmed, obtain the traffic sign classification that the weight of normalization layer output is maximum; Be identify the traffic sign classification obtained by traffic sign category setting maximum for weight.
Fig. 6 shows the topology example figure of the device of the identification traffic sign according to the embodiment of the present application.
As shown in Figure 6, the device 600 of described identification traffic sign can comprise: characteristic value acquisition module 610, road traffic sign detection module 620 and Traffic Sign Recognition module 630.
Characteristic value acquisition module 610, divides by panorama spherical diagram picture the eigenwert that the scanning window image that obtains obtained by predetermined characteristic algorithm at predetermined integral passage for obtaining.
Alternatively, characteristic value acquisition module 610 can comprise: integrogram obtains the first submodule 611, for obtaining the integrogram being divided the scanning window image obtained by panorama spherical diagram picture at the integrogram of predetermined integral passage; And eigenwert obtains the first submodule 612, for the integrogram according to scanning window image, obtain the eigenwert of scanning window image.
Road traffic sign detection module 620, for according to the eigenwert of scanning window image and the detection sorter model of training in advance, detect scanning window image, obtain traffic sign video in window to be confirmed, wherein, the detection sorter model of training in advance obtains according to the sample of scanning window image and eigenwert training thereof.
Traffic Sign Recognition module 630, for the convolutional neural networks model according to training in advance, identify traffic sign video in window to be confirmed, obtain traffic sign classification, wherein, the convolutional neural networks model of training in advance obtains according to the sample of confirmed traffic sign video in window and the training of traffic sign classification thereof.
The device of the identification traffic sign of the above embodiments of the present application, improves the accuracy of detection and Identification traffic sign in panoramic picture, improves the renewal efficiency of road net data simultaneously.
Alternatively, the detection sorter model of above-mentioned training in advance can obtain by the following method: obtain the positive sample in the sample of scanning window image and negative sample, wherein positive sample comprises the video in window of traffic sign or comprises the video in window of traffic sign and surrounding expansion presetted pixel thereof, and negative sample comprises the scanning window image removing positive sample; Obtain the eigenwert that positive sample and negative sample are obtained by predetermined characteristic algorithm at predetermined integral passage; According to positive sample and negative sample and the eigenwert that obtained by predetermined characteristic algorithm at predetermined integral passage thereof, detect sorter model by boosting Algorithm for Training, obtain the detection sorter model of training in advance.
The method of the detection sorter model of above-mentioned acquisition training in advance, can be realized by the detection sorter model trainer shown in Fig. 7.
Fig. 7 shows the exemplary block diagram of the detection sorter model trainer according to the embodiment of the present application.
As shown in Figure 7, detect sorter model trainer 700 can comprise: sample acquisition module 701, sample characteristics acquisition module 702 and detection model training module 703.
Sample acquisition module 701, for obtaining positive sample in the sample of scanning window image and negative sample, wherein positive sample comprises the video in window of traffic sign or comprises the video in window of traffic sign and surrounding expansion presetted pixel thereof, and negative sample comprises the scanning window image removing positive sample.
Sample characteristics acquisition module 702, for obtaining the eigenwert that positive sample and negative sample are obtained by predetermined characteristic algorithm at predetermined integral passage.
Wherein, predetermined integral passage comprises following one or more: gray level image passage, red, green, blue monochrome image passage, the gradient image passage of different angles parameter, and edge-detected image passage; And predetermined characteristic algorithm comprises following one or more: Lis Hartel levies algorithm, local binary patterns characteristics algorithm, histogram feature algorithm and random to characteristics algorithm.
Detection model training module 703, for according to positive sample and negative sample and the eigenwert that obtained by predetermined characteristic algorithm at predetermined integral passage thereof, is detected sorter model by boosting Algorithm for Training, obtains the detection sorter model of training in advance.
Below in conjunction with Fig. 8, on the basis of above-described embodiment, the device obtaining traffic sign video in window to be confirmed is described.
Fig. 8 shows the exemplary block diagram of the device for obtaining traffic sign video in window to be confirmed.
As shown in Figure 8, the device 800 for obtaining traffic sign video in window to be confirmed can comprise: down-sampled submodule 801, integrogram obtain the second submodule 802, eigenwert second obtains submodule 803 and multistage detection sub-module 804.
Down-sampled submodule 801, for down-sampled continuously to panorama spherical diagram picture, obtains image pyramid.
Integrogram obtains the second submodule 802, for obtaining the integrogram being divided the scanning window image of every grade of image in the image pyramid obtained by image pyramid at the integrogram of predetermined integral passage.
Eigenwert second obtains submodule 803, for the integrogram of the scanning window image according to every grade of image, obtains the eigenwert of the scanning window image of every grade of image.
Multistage detection sub-module 804, for the eigenwert of the scanning window image according to every grade of image and the detection sorter model of training in advance, detects the scanning window image of every grade of image, obtains traffic sign video in window to be confirmed.
By the device of the above-mentioned acquisition of the embodiment of the present application traffic sign video in window to be confirmed, the accuracy obtaining traffic sign video in window to be confirmed and the speed improving detection traffic sign video in window to be confirmed can be improved further.
Alternatively, the convolutional neural networks model of the training in advance in above-mentioned Fig. 6 can obtain by the following method: according to Gaussian distribution, the convolutional layer of the convolutional neural networks model that initialization is preset and the weight of full articulamentum, wherein, described default convolutional neural networks model comprises the convolutional layer, abstraction, layer, full articulamentum and the normalization layer that connect successively; According to sample and the traffic sign classification thereof of confirmed traffic sign video in window, iteration is carried out by the weight of error back propagation BP algorithm to described convolutional layer and described full articulamentum, if the difference of the weight after the weight after current iteration and last iteration is less than preset value, then the weight after current iteration being defined as optimal weights, is the convolutional neural networks model of described training in advance by the convolutional neural networks model specification comprising optimal weights; If or there is the weight after the iteration that error rate is minimum, then the weight after iteration minimum for error rate being defined as optimal weights, is the convolutional neural networks model of described training in advance by the convolutional neural networks model specification comprising optimal weights.
Wherein, the sample of confirmed traffic sign video in window is in the sample extracted in traffic sign video in window to be confirmed, the sample comprising the video in window of traffic sign after confirming according to the artificial mark received.
Further, the sample of confirmed traffic sign video in window can comprise: the original sample having confirmed traffic sign video in window, and has confirmed that the sample of traffic sign video in window to carry out after following one or more process normalizing to the image of preset window size by original: rotate, Pan and Zoom.
Wherein, the convolutional layer connected successively, abstraction, layer, full articulamentum and normalization layer comprise: multiple convolutional layer and with convolutional layer abstraction, layer, more than one full articulamentum and a normalization layer one to one.
Further, the convolutional neural networks model preset can also comprise loss function layer.
Correspond to the convolutional neural networks model preset and also comprise loss function layer, carrying out iteration by the weight of error back propagation BP algorithm to described convolutional layer and described full articulamentum can comprise: by loss function and BP algorithm, the weight to convolutional layer and full articulamentum carries out iteration respectively.
The method of the convolutional neural networks model of above-mentioned acquisition training in advance can be realized by the device of the convolutional neural networks model of the training training in advance shown in Fig. 9.
Fig. 9 shows the exemplary block diagram of the device of the convolutional neural networks model for training training in advance according to the embodiment of the present application.
As shown in Figure 9, the device 900 for the convolutional neural networks model of training training in advance comprises:
Initialization module 901, for according to Gaussian distribution, the convolutional layer of the convolutional neural networks model that initialization is preset and the weight of full articulamentum, wherein, the convolutional neural networks model preset comprises the convolutional layer, abstraction, layer, full articulamentum and the normalization layer that connect successively.
Wherein, the convolutional layer connected successively, abstraction, layer, full articulamentum and normalization layer comprise: multiple convolutional layer and with convolutional layer abstraction, layer, more than one full articulamentum and a normalization layer one to one.
Alternatively, the convolutional neural networks model preset also comprises loss function layer.
Weight iteration module 902, for according to the sample of confirmed traffic sign video in window and traffic sign classification thereof, carries out iteration by the weight of error back propagation BP algorithm to convolutional layer and full articulamentum;
First optimal weights determination module 903, if be less than preset value for the difference of the weight after the weight after current iteration and last iteration, is then defined as optimal weights by the weight after current iteration.
Second optimal weights determination module 904, if for there is the weight after the iteration that error rate is minimum, be then defined as optimal weights by the weight after iteration minimum for error rate.
Model specification module 905, for by the convolutional neural networks model specification comprising optimal weights being the convolutional neural networks model of training in advance.
Correspond to the convolutional neural networks model preset and also comprise loss function layer, weight determination module 902 is further used for the weight to convolutional layer and full articulamentum by loss function and BP algorithm and carries out iteration respectively.
After the device 900 of the convolutional neural networks model for training training in advance trains the convolutional neural networks model of training in advance, Traffic Sign Recognition module 630 then in Fig. 6 can comprise: weight limit identification module (not shown), for by traffic sign video in window input convolutional neural networks model to be confirmed, obtain the traffic sign classification that the weight of normalization layer output is maximum; And flag category setting module (not shown), for by traffic sign category setting maximum for weight being the traffic sign class that identification obtains.
Should be appreciated that all unit recorded in device 600 are corresponding with each step in the method described with reference to figure 1.The all unit recorded in device 700 are corresponding with each step in the method described with reference to figure 2.The all unit recorded in device 800 are corresponding with each step in the method described with reference to figure 3.The all unit recorded in device 900 are corresponding with each step in the method described with reference to figure 4.Thus, above for identifying the operation that the method for traffic sign describes and the unit that feature is equally applicable to device 600 and wherein comprises, above for the unit that operation and the feature of the method description of the detection model of training training in advance are equally applicable to device 700 and wherein comprise, above for the unit that operation and the feature of the method description of acquisition traffic sign video in window to be confirmed are equally applicable to device 800 and wherein comprise, above for the unit that operation and the feature of the method description of the convolutional neural networks model of training training in advance are equally applicable to device 900 and wherein comprise, do not repeat them here.Corresponding units in device 600,700,800 and 900 can cooperatively interact the scheme realizing the embodiment of the present application with the unit in terminal device and/or server.
Be described in module involved in the embodiment of the present application to be realized by the mode of software, also can be realized by the mode of hardware.Described module also can be arranged within a processor, such as, can be described as: a kind of processor comprises characteristic value acquisition module, detection module and identification module.Wherein, the title of these modules does not form the restriction to this module itself under certain conditions, such as, characteristic value acquisition module can also be described to " for obtaining the module being divided the eigenwert that the scanning window image that obtains is obtained by predetermined characteristic algorithm at predetermined integral passage by panorama spherical diagram picture ".
As another aspect, present invention also provides a kind of computer-readable recording medium, this computer-readable recording medium can be the computer-readable recording medium comprised in device described in above-described embodiment; Also can be individualism, be unkitted the computer-readable recording medium allocated in terminal.Described computer-readable recording medium stores more than one or one program, and described program is used for performance description in the method for the identification traffic sign of the application by one or more than one processor.
More than describe and be only the preferred embodiment of the application and the explanation to institute's application technology principle.Those skilled in the art are to be understood that, invention scope involved in the application, be not limited to the technical scheme of the particular combination of above-mentioned technical characteristic, also should be encompassed in when not departing from described inventive concept, other technical scheme of being carried out combination in any by above-mentioned technical characteristic or its equivalent feature and being formed simultaneously.The technical characteristic that such as, disclosed in above-mentioned feature and the application (but being not limited to) has similar functions is replaced mutually and the technical scheme formed.
Claims (14)
1. identify a method for traffic sign, it is characterized in that, comprising:
Obtain and divide by panorama spherical diagram picture the eigenwert that the scanning window image that obtains obtained by predetermined characteristic algorithm at predetermined integral passage;
According to the eigenwert of described scanning window image and the detection sorter model of training in advance, detect scanning window image, obtain traffic sign video in window to be confirmed, wherein, the detection sorter model of described training in advance obtains according to the sample of scanning window image and eigenwert training thereof;
According to the convolutional neural networks model of training in advance, identify traffic sign video in window to be confirmed, obtain traffic sign classification, wherein, the convolutional neural networks model of described training in advance obtains according to the sample of confirmed traffic sign video in window and the training of traffic sign classification thereof.
2. method according to claim 1, is characterized in that, obtains the detection sorter model of described training in advance by the following method:
Obtain the positive sample in the sample of scanning window image and negative sample, wherein said positive sample comprises the video in window of traffic sign or comprises the video in window of traffic sign and surrounding expansion presetted pixel thereof, and described negative sample comprises the scanning window image removing positive sample;
Obtain the eigenwert that described positive sample and negative sample are obtained by predetermined characteristic algorithm at predetermined integral passage;
According to described positive sample and negative sample and the eigenwert that obtained by predetermined characteristic algorithm at predetermined integral passage thereof, detect sorter model by boosting Algorithm for Training, obtain the detection sorter model of described training in advance.
3. method according to claim 1, is characterized in that, described acquisition divides by panorama spherical diagram picture the eigenwert that the scanning window image that obtains obtained by predetermined characteristic algorithm at predetermined integral passage and comprises:
Obtain the integrogram being divided the scanning window image obtained by panorama spherical diagram picture at the integrogram of predetermined integral passage;
According to the integrogram of described scanning window image, obtain the eigenwert of described scanning window image.
4. method according to claim 3, it is characterized in that, the integrogram that described acquisition divides at the integrogram of predetermined integral passage the scanning window image obtained by panorama spherical diagram picture comprises: down-sampled continuously to described panorama spherical diagram picture, obtains image pyramid; Obtain the integrogram being divided the scanning window image of every grade of image in the image pyramid obtained by image pyramid at the integrogram of predetermined integral passage;
The described integrogram according to described scanning window image, the eigenwert obtaining described scanning window image comprises: according to the integrogram of the scanning window image of described every grade of image, obtains the eigenwert of the scanning window image of every grade of image; And
Described according to the eigenwert of described scanning window image and the detection sorter model of training in advance, detect scanning window image, obtain traffic sign video in window to be confirmed to comprise: according to the eigenwert of the scanning window image of described every grade of image and the detection sorter model of training in advance, detect the scanning window image of described every grade of image, obtain traffic sign video in window to be confirmed.
5. according to the method one of claim 1-4 Suo Shu, it is characterized in that, obtain the convolutional neural networks model of described training in advance by the following method:
According to Gaussian distribution, the convolutional layer of the convolutional neural networks model that initialization is preset and the weight of full articulamentum, wherein, described default convolutional neural networks model comprises the convolutional layer, abstraction, layer, full articulamentum and the normalization layer that connect successively;
According to sample and the traffic sign classification thereof of confirmed traffic sign video in window, iteration is carried out by the weight of error back propagation BP algorithm to described convolutional layer and described full articulamentum, if the difference of the weight after the weight after current iteration and last iteration is less than preset value, then the weight after current iteration being defined as optimal weights, is the convolutional neural networks model of described training in advance by the convolutional neural networks model specification comprising optimal weights; If or there is the weight after the iteration that error rate is minimum, then the weight after iteration minimum for error rate being defined as optimal weights, is the convolutional neural networks model of described training in advance by the convolutional neural networks model specification comprising optimal weights.
6. method according to claim 5, is characterized in that, it is characterized in that, described default convolutional neural networks model also comprises loss function layer; And
Describedly carry out iteration by the weight of error back propagation BP algorithm to described convolutional layer and described full articulamentum and comprise: the weight to described convolutional layer and described full articulamentum carries out iteration respectively by loss function and BP algorithm.
7. method according to claim 5, is characterized in that, the described convolutional neural networks model according to training in advance, identifies traffic sign video in window to be confirmed, obtains traffic sign classification and comprises:
Described traffic sign video in window to be confirmed is inputted described convolutional neural networks model, obtains the traffic sign classification that the weight of normalization layer output is maximum;
Be identify the traffic sign classification obtained by traffic sign category setting maximum for described weight.
8. identify a device for traffic sign, it is characterized in that, comprising:
Characteristic value acquisition module, divides by panorama spherical diagram picture the eigenwert that the scanning window image that obtains obtained by predetermined characteristic algorithm at predetermined integral passage for obtaining;
Road traffic sign detection module, for according to the eigenwert of described scanning window image and the detection sorter model of training in advance, detect scanning window image, obtain traffic sign video in window to be confirmed, wherein, the detection sorter model of described training in advance obtains according to the sample of scanning window image and eigenwert training thereof;
Traffic Sign Recognition module, for the convolutional neural networks model according to training in advance, identify traffic sign video in window to be confirmed, obtain traffic sign classification, wherein, the convolutional neural networks model of described training in advance obtains according to the sample of confirmed traffic sign video in window and the training of traffic sign classification thereof.
9. device according to claim 8, is characterized in that, the detection sorter model of described training in advance obtains by the following method:
Obtain the positive sample in the sample of scanning window image and negative sample, wherein said positive sample comprises the video in window of traffic sign or comprises the video in window of traffic sign and surrounding expansion presetted pixel thereof, and described negative sample comprises the scanning window image removing positive sample;
Obtain the eigenwert that described positive sample and negative sample are obtained by predetermined characteristic algorithm at predetermined integral passage;
According to described positive sample and negative sample and the eigenwert that obtained by predetermined characteristic algorithm at predetermined integral passage thereof, detect sorter model by boosting Algorithm for Training, obtain the detection sorter model of described training in advance.
10. device according to claim 8, is characterized in that, described characteristic value acquisition module comprises:
Integrogram obtains the first submodule, for obtaining the integrogram being divided the scanning window image obtained by panorama spherical diagram picture at the integrogram of predetermined integral passage;
Eigenwert obtains the first submodule, for the integrogram according to described scanning window image, obtains the eigenwert of described scanning window image.
11. devices according to claim 10, is characterized in that, described integrogram obtains the first submodule and comprises: down-sampled submodule, for down-sampled continuously to described panorama spherical diagram picture, obtain image pyramid; Integrogram obtains the second submodule, for obtaining the integrogram being divided the scanning window image of every grade of image in the image pyramid obtained by image pyramid at the integrogram of predetermined integral passage;
Described eigenwert obtains the first submodule and comprises: eigenwert second obtains submodule, for the integrogram of the scanning window image according to described every grade of image, obtains the eigenwert of the scanning window image of every grade of image; And
Described detection module comprises: multistage detection sub-module, for the eigenwert of the scanning window image according to described every grade of image and the detection sorter model of training in advance, detect the scanning window image of described every grade of image, obtain traffic sign video in window to be confirmed.
12. one of-11 described devices according to Claim 8, it is characterized in that, the convolutional neural networks model of described training in advance obtains by the following method:
According to Gaussian distribution, the convolutional layer of the convolutional neural networks model that initialization is preset and the weight of full articulamentum, wherein, described default convolutional neural networks model comprises the convolutional layer, abstraction, layer, full articulamentum and the normalization layer that connect successively;
According to sample and the traffic sign classification thereof of confirmed traffic sign video in window, iteration is carried out by the weight of error back propagation BP algorithm to described convolutional layer and described full articulamentum, if the difference of the weight after the weight after current iteration and last iteration is less than preset value, then the weight after current iteration being defined as optimal weights, is the convolutional neural networks model of described training in advance by the convolutional neural networks model specification comprising optimal weights; If or there is the weight after the iteration that error rate is minimum, then the weight after iteration minimum for error rate being defined as optimal weights, is the convolutional neural networks model of described training in advance by the convolutional neural networks model specification comprising optimal weights.
13. devices according to claim 12, is characterized in that, it is characterized in that, described default convolutional neural networks model also comprises loss function layer; And
Described weight determination module is further used for the weight to described convolutional layer and described full articulamentum by loss function and BP algorithm and carries out iteration respectively.
14. devices according to claim 12, is characterized in that, described Traffic Sign Recognition module comprises:
Weight limit identification module, for described traffic sign video in window to be confirmed is inputted described convolutional neural networks model, obtains the traffic sign classification that the weight of normalization layer output is maximum;
Flag category setting module, for by traffic sign category setting maximum for described weight being the traffic sign class that identification obtains.
Priority Applications (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510150525.8A CN104700099B (en) | 2015-03-31 | 2015-03-31 | The method and apparatus for recognizing traffic sign |
| JP2016559945A JP6400117B2 (en) | 2015-03-31 | 2015-12-25 | Method and apparatus for recognizing traffic signs |
| KR1020167027213A KR101856584B1 (en) | 2015-03-31 | 2015-12-25 | Method and device for identifying traffic signs |
| PCT/CN2015/098903 WO2016155371A1 (en) | 2015-03-31 | 2015-12-25 | Method and device for recognizing traffic signs |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510150525.8A CN104700099B (en) | 2015-03-31 | 2015-03-31 | The method and apparatus for recognizing traffic sign |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104700099A true CN104700099A (en) | 2015-06-10 |
| CN104700099B CN104700099B (en) | 2017-08-11 |
Family
ID=53347198
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510150525.8A Active CN104700099B (en) | 2015-03-31 | 2015-03-31 | The method and apparatus for recognizing traffic sign |
Country Status (4)
| Country | Link |
|---|---|
| JP (1) | JP6400117B2 (en) |
| KR (1) | KR101856584B1 (en) |
| CN (1) | CN104700099B (en) |
| WO (1) | WO2016155371A1 (en) |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105335754A (en) * | 2015-10-29 | 2016-02-17 | 小米科技有限责任公司 | Character recognition method and device |
| CN105678322A (en) * | 2015-12-31 | 2016-06-15 | 百度在线网络技术（北京）有限公司 | Sample labeling method and apparatus |
| CN105869016A (en) * | 2016-03-28 | 2016-08-17 | 天津中科智能识别产业技术研究院有限公司 | Method for estimating click through rate based on convolution neural network |
| CN105930830A (en) * | 2016-05-18 | 2016-09-07 | 大连理工大学 | A road traffic sign recognition method based on convolutional neural network |
| CN105956524A (en) * | 2016-04-22 | 2016-09-21 | 北京智芯原动科技有限公司 | Method and device for identifying traffic signs |
| WO2016155371A1 (en) * | 2015-03-31 | 2016-10-06 | 百度在线网络技术（北京）有限公司 | Method and device for recognizing traffic signs |
| CN106022380A (en) * | 2016-05-25 | 2016-10-12 | 中国科学院自动化研究所 | Individual identity identification method based on deep learning |
| CN106778835A (en) * | 2016-11-29 | 2017-05-31 | 武汉大学 | The airport target by using remote sensing image recognition methods of fusion scene information and depth characteristic |
| CN107067043A (en) * | 2017-05-25 | 2017-08-18 | 哈尔滨工业大学 | A kind of diseases and pests of agronomic crop detection method |
| CN107679577A (en) * | 2017-10-12 | 2018-02-09 | 理光图像技术（上海）有限公司 | image detecting method based on deep learning |
| CN107944437A (en) * | 2017-12-31 | 2018-04-20 | 广州二元科技有限公司 | A kind of Face detection method based on neutral net and integral image |
| CN105069472B (en) * | 2015-08-03 | 2018-07-27 | 电子科技大学 | A kind of vehicle checking method adaptive based on convolutional neural networks |
| CN108520212A (en) * | 2018-03-27 | 2018-09-11 | 东华大学 | Method for traffic sign detection based on improved convolutional neural networks |
| CN108549927A (en) * | 2018-03-16 | 2018-09-18 | 深圳市商汤科技有限公司 | The training method and feature of object detection network supervise drawing generating method |
| CN108628206A (en) * | 2017-03-17 | 2018-10-09 | 通用汽车环球科技运作有限责任公司 | Road construction detecting system and method |
| WO2018188270A1 (en) * | 2017-04-14 | 2018-10-18 | 北京图森未来科技有限公司 | Image semantic segmentation method and device |
| CN108960308A (en) * | 2018-06-25 | 2018-12-07 | 中国科学院自动化研究所 | Traffic sign recognition method, device, car-mounted terminal and vehicle |
| CN109102017A (en) * | 2018-08-09 | 2018-12-28 | 百度在线网络技术（北京）有限公司 | Neural network model processing method, device, equipment and readable storage medium storing program for executing |
| CN109492628A (en) * | 2018-10-08 | 2019-03-19 | 杭州电子科技大学 | A kind of implementation method of the spherical positioning system applied to the undisciplined crawl in classroom |
| CN109726728A (en) * | 2017-10-31 | 2019-05-07 | 高德软件有限公司 | A kind of training data generation method and device |
| CN105930830B (en) * | 2016-05-18 | 2019-07-16 | 大连理工大学 | A road traffic sign recognition method based on convolutional neural network |
| CN110096937A (en) * | 2018-01-31 | 2019-08-06 | 北京四维图新科技股份有限公司 | A kind of method and device of the image recognition for assisting Vehicular automatic driving |
| CN110163140A (en) * | 2019-05-15 | 2019-08-23 | 腾讯科技（深圳）有限公司 | Crowd density picture capturing method and device |
| CN110287756A (en) * | 2018-03-19 | 2019-09-27 | 株式会社东芝 | Identification devices and vehicle systems |
| CN110796109A (en) * | 2019-11-05 | 2020-02-14 | 哈尔滨理工大学 | Driver distraction behavior identification method based on model fusion |
| CN111178371A (en) * | 2019-12-17 | 2020-05-19 | 深圳市优必选科技股份有限公司 | Object detection method, device and computer storage medium |
| CN111488782A (en) * | 2019-01-25 | 2020-08-04 | 斯特拉德视觉公司 | Method and device for detecting road barrier by using Laplacian pyramid network |
| CN112085056A (en) * | 2020-08-05 | 2020-12-15 | 深圳市优必选科技股份有限公司 | Object detection model generation method, device, device and storage medium |
| CN115131745A (en) * | 2022-06-09 | 2022-09-30 | 合肥讯图信息科技有限公司 | Traffic signal lamp identification method and system under night scene |
| CN119649333A (en) * | 2024-09-18 | 2025-03-18 | 海南省工业研究所 | Traffic sign detection method for complex scenes of autonomous driving based on multi-channel image extraction |
| CN120318782A (en) * | 2025-06-12 | 2025-07-15 | 武汉大学 | Ultra-wide-angle traffic sign detection method and system based on adaptive dynamic pruning |
Families Citing this family (36)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11308391B2 (en) * | 2017-03-06 | 2022-04-19 | Baidu Usa Llc | Offline combination of convolutional/deconvolutional and batch-norm layers of convolutional neural network models for autonomous driving vehicles |
| KR101834778B1 (en) | 2017-05-30 | 2018-03-07 | 만도헬라일렉트로닉스(주) | Apparatus for recognizing traffic sign and method thereof |
| WO2019009447A1 (en) * | 2017-07-06 | 2019-01-10 | 삼성전자 주식회사 | Method for encoding/decoding image and device therefor |
| US10986356B2 (en) | 2017-07-06 | 2021-04-20 | Samsung Electronics Co., Ltd. | Method for encoding/decoding image and device therefor |
| CN110337813B (en) | 2017-07-06 | 2023-04-18 | 三星电子株式会社 | Method for encoding/decoding image and apparatus therefor |
| CN109255279A (en) * | 2017-07-13 | 2019-01-22 | 深圳市凯立德科技股份有限公司 | A kind of method and system of road traffic sign detection identification |
| US9947103B1 (en) * | 2017-10-03 | 2018-04-17 | StradVision, Inc. | Learning method and learning device for improving image segmentation and testing method and testing device using the same |
| KR102102161B1 (en) * | 2018-05-18 | 2020-04-20 | 오드컨셉 주식회사 | Method, apparatus and computer program for extracting representative feature of object in image |
| KR102106029B1 (en) * | 2018-07-19 | 2020-05-04 | 네이버랩스 주식회사 | Method and system for improving signage detection performance |
| KR102096926B1 (en) * | 2018-07-19 | 2020-04-06 | 네이버랩스 주식회사 | Method and system for detecting change point of interest |
| CN113610923B (en) * | 2018-07-26 | 2025-07-22 | 上海联影智能医疗科技有限公司 | Scanning positioning method, scanning positioning device, computer equipment and computer readable storage medium |
| CN110879950A (en) * | 2018-09-06 | 2020-03-13 | 北京市商汤科技开发有限公司 | Multi-stage target classification and traffic sign detection method and device, equipment and medium |
| KR102525576B1 (en) | 2018-10-19 | 2023-04-26 | 삼성전자주식회사 | Apparatus and method for performing artificial intelligence encoding and artificial intelligence decoding of image |
| WO2020080765A1 (en) | 2018-10-19 | 2020-04-23 | Samsung Electronics Co., Ltd. | Apparatuses and methods for performing artificial intelligence encoding and artificial intelligence decoding on image |
| CN109919017B (en) * | 2019-01-28 | 2023-09-01 | 平安科技（深圳）有限公司 | Face recognition optimization method, device, computer equipment and storage medium |
| CN110175500B (en) * | 2019-04-03 | 2024-01-19 | 平安科技（深圳）有限公司 | Finger vein comparison method, device, computer equipment and storage medium |
| CN113409393B (en) * | 2019-05-17 | 2023-10-03 | 百度在线网络技术（北京）有限公司 | Method and device for identifying traffic sign |
| CN110348396B (en) * | 2019-07-15 | 2022-02-11 | 南京信息工程大学 | Deep learning-based method and device for recognizing character traffic signs above roads |
| KR102436512B1 (en) | 2019-10-29 | 2022-08-25 | 삼성전자주식회사 | Method and Apparatus for video encoding and Method and Apparatus for video decoding |
| CN110889378B (en) * | 2019-11-28 | 2023-06-09 | 湖南率为控制科技有限公司 | A multi-view fusion traffic sign detection and recognition method and system |
| CN111062885B (en) * | 2019-12-09 | 2023-09-12 | 中国科学院自动化研究所 | Sign detection model training and sign detection method based on multi-stage transfer learning |
| CN111178153A (en) * | 2019-12-09 | 2020-05-19 | 武汉光庭信息技术股份有限公司 | Traffic sign detection method and system |
| CN111199217B (en) * | 2020-01-09 | 2023-03-28 | 上海应用技术大学 | Traffic sign identification method and system based on convolutional neural network |
| KR102287942B1 (en) | 2020-02-24 | 2021-08-09 | 삼성전자주식회사 | Apparatus and method for performing artificial intelligence encoding and artificial intelligence decoding of image using pre-processing |
| CN111310844B (en) * | 2020-02-26 | 2022-09-16 | 广州华工邦元信息技术有限公司 | Vehicle identification model construction method and device and identification method and device |
| KR102189926B1 (en) * | 2020-03-30 | 2020-12-11 | 네이버랩스 주식회사 | Method and system for detecting change point of interest |
| CN111597875A (en) * | 2020-04-01 | 2020-08-28 | 吉利汽车研究院（宁波）有限公司 | Traffic sign identification method, device, equipment and storage medium |
| CN111881944B (en) * | 2020-07-08 | 2024-09-17 | 贵州无忧天空科技有限公司 | Image identification method, electronic device and computer readable medium |
| CN112149528A (en) * | 2020-09-08 | 2020-12-29 | 东易日盛家居装饰集团股份有限公司 | Panorama target detection method, system, medium and equipment |
| CN113033471A (en) * | 2021-04-15 | 2021-06-25 | 北京百度网讯科技有限公司 | Traffic abnormality detection method, apparatus, device, storage medium, and program product |
| CN113850195B (en) * | 2021-09-27 | 2024-05-28 | 新讯数字科技(杭州)有限公司 | An AI intelligent object recognition method based on 3D vision |
| CN115063779A (en) * | 2022-06-28 | 2022-09-16 | 南京信息工程大学 | Traffic sign detection method and device combined with image mask and storage medium |
| CN116524455A (en) * | 2023-03-02 | 2023-08-01 | 大连民族大学 | Night traffic sign recognition method, system and storage medium based on deep learning |
| CN116301647B (en) | 2023-03-24 | 2025-12-23 | 昆仑芯(北京)科技有限公司 | Electronic devices, methods for determining memory access efficiency, and storage media |
| CN116486369A (en) * | 2023-04-06 | 2023-07-25 | 清华大学 | A traffic sign detection method, device, electronic equipment and storage medium |
| CN118521997A (en) * | 2024-07-25 | 2024-08-20 | 比亚迪股份有限公司 | Driving perception method and device, electronic equipment and vehicle |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130272575A1 (en) * | 2011-11-01 | 2013-10-17 | Intel Corporation | Object detection using extended surf features |
| CN104036323A (en) * | 2014-06-26 | 2014-09-10 | 叶茂 | Vehicle detection method based on convolutional neural network |
Family Cites Families (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH08221378A (en) * | 1995-02-10 | 1996-08-30 | Ricoh Co Ltd | Learning machine |
| JP2004191339A (en) * | 2002-12-13 | 2004-07-08 | Sharp Corp | Position information search method, position information search device, position information search terminal, and position information search system |
| JP2005311691A (en) * | 2004-04-21 | 2005-11-04 | Toyota Central Res & Dev Lab Inc | Object detection apparatus and method |
| JP5364578B2 (en) * | 2006-07-12 | 2013-12-11 | コファックス， インコーポレイテッド | Method and system for transductive data classification and data classification method using machine learning technique |
| JP2013196454A (en) * | 2012-03-21 | 2013-09-30 | Honda Elesys Co Ltd | Image processor, image processing method and image processing program |
| CN103366177B (en) * | 2012-03-28 | 2016-12-07 | 佳能株式会社 | Object detection classifier generation method and equipment, image object detection method and equipment |
| JP2014010633A (en) * | 2012-06-29 | 2014-01-20 | Honda Elesys Co Ltd | Image recognition device, image recognition method, and image recognition program |
| JP2014071832A (en) * | 2012-10-01 | 2014-04-21 | Toshiba Corp | Object detection apparatus and detection method of the same |
| CN104700099B (en) * | 2015-03-31 | 2017-08-11 | 百度在线网络技术（北京）有限公司 | The method and apparatus for recognizing traffic sign |
-
2015
- 2015-03-31 CN CN201510150525.8A patent/CN104700099B/en active Active
- 2015-12-25 WO PCT/CN2015/098903 patent/WO2016155371A1/en not_active Ceased
- 2015-12-25 JP JP2016559945A patent/JP6400117B2/en not_active Expired - Fee Related
- 2015-12-25 KR KR1020167027213A patent/KR101856584B1/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130272575A1 (en) * | 2011-11-01 | 2013-10-17 | Intel Corporation | Object detection using extended surf features |
| CN104036323A (en) * | 2014-06-26 | 2014-09-10 | 叶茂 | Vehicle detection method based on convolutional neural network |
Cited By (41)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016155371A1 (en) * | 2015-03-31 | 2016-10-06 | 百度在线网络技术（北京）有限公司 | Method and device for recognizing traffic signs |
| CN105069472B (en) * | 2015-08-03 | 2018-07-27 | 电子科技大学 | A kind of vehicle checking method adaptive based on convolutional neural networks |
| CN105335754A (en) * | 2015-10-29 | 2016-02-17 | 小米科技有限责任公司 | Character recognition method and device |
| CN105678322A (en) * | 2015-12-31 | 2016-06-15 | 百度在线网络技术（北京）有限公司 | Sample labeling method and apparatus |
| CN105869016A (en) * | 2016-03-28 | 2016-08-17 | 天津中科智能识别产业技术研究院有限公司 | Method for estimating click through rate based on convolution neural network |
| CN105956524A (en) * | 2016-04-22 | 2016-09-21 | 北京智芯原动科技有限公司 | Method and device for identifying traffic signs |
| CN105930830A (en) * | 2016-05-18 | 2016-09-07 | 大连理工大学 | A road traffic sign recognition method based on convolutional neural network |
| CN105930830B (en) * | 2016-05-18 | 2019-07-16 | 大连理工大学 | A road traffic sign recognition method based on convolutional neural network |
| CN106022380A (en) * | 2016-05-25 | 2016-10-12 | 中国科学院自动化研究所 | Individual identity identification method based on deep learning |
| CN106778835A (en) * | 2016-11-29 | 2017-05-31 | 武汉大学 | The airport target by using remote sensing image recognition methods of fusion scene information and depth characteristic |
| CN106778835B (en) * | 2016-11-29 | 2020-03-24 | 武汉大学 | Remote sensing image airport target identification method fusing scene information and depth features |
| CN108628206A (en) * | 2017-03-17 | 2018-10-09 | 通用汽车环球科技运作有限责任公司 | Road construction detecting system and method |
| US11875511B2 (en) | 2017-04-14 | 2024-01-16 | Beijing Tusen Zhitu Technology Co., Ltd. | Method and device for semantic segmentation of image |
| US11205271B2 (en) | 2017-04-14 | 2021-12-21 | Beijing Tusen Zhitu Technology Co., Ltd. | Method and device for semantic segmentation of image |
| WO2018188270A1 (en) * | 2017-04-14 | 2018-10-18 | 北京图森未来科技有限公司 | Image semantic segmentation method and device |
| CN107067043B (en) * | 2017-05-25 | 2020-07-24 | 哈尔滨工业大学 | Crop disease and insect pest detection method |
| CN107067043A (en) * | 2017-05-25 | 2017-08-18 | 哈尔滨工业大学 | A kind of diseases and pests of agronomic crop detection method |
| CN107679577A (en) * | 2017-10-12 | 2018-02-09 | 理光图像技术（上海）有限公司 | image detecting method based on deep learning |
| CN109726728A (en) * | 2017-10-31 | 2019-05-07 | 高德软件有限公司 | A kind of training data generation method and device |
| CN107944437A (en) * | 2017-12-31 | 2018-04-20 | 广州二元科技有限公司 | A kind of Face detection method based on neutral net and integral image |
| CN110096937A (en) * | 2018-01-31 | 2019-08-06 | 北京四维图新科技股份有限公司 | A kind of method and device of the image recognition for assisting Vehicular automatic driving |
| CN108549927A (en) * | 2018-03-16 | 2018-09-18 | 深圳市商汤科技有限公司 | The training method and feature of object detection network supervise drawing generating method |
| CN110287756A (en) * | 2018-03-19 | 2019-09-27 | 株式会社东芝 | Identification devices and vehicle systems |
| CN108520212A (en) * | 2018-03-27 | 2018-09-11 | 东华大学 | Method for traffic sign detection based on improved convolutional neural networks |
| CN108960308A (en) * | 2018-06-25 | 2018-12-07 | 中国科学院自动化研究所 | Traffic sign recognition method, device, car-mounted terminal and vehicle |
| CN109102017B (en) * | 2018-08-09 | 2021-08-03 | 百度在线网络技术（北京）有限公司 | Neural network model processing method, device, equipment and readable storage medium |
| CN109102017A (en) * | 2018-08-09 | 2018-12-28 | 百度在线网络技术（北京）有限公司 | Neural network model processing method, device, equipment and readable storage medium storing program for executing |
| CN109492628A (en) * | 2018-10-08 | 2019-03-19 | 杭州电子科技大学 | A kind of implementation method of the spherical positioning system applied to the undisciplined crawl in classroom |
| CN109492628B (en) * | 2018-10-08 | 2021-11-16 | 杭州电子科技大学 | Implementation method of spherical positioning system applied to classroom discipline grabbing |
| CN111488782B (en) * | 2019-01-25 | 2023-04-07 | 斯特拉德视觉公司 | Method and device for detecting road barrier by using Laplacian pyramid network |
| CN111488782A (en) * | 2019-01-25 | 2020-08-04 | 斯特拉德视觉公司 | Method and device for detecting road barrier by using Laplacian pyramid network |
| CN110163140B (en) * | 2019-05-15 | 2024-09-06 | 腾讯科技（深圳）有限公司 | Crowd density map acquisition method and device |
| CN110163140A (en) * | 2019-05-15 | 2019-08-23 | 腾讯科技（深圳）有限公司 | Crowd density picture capturing method and device |
| CN110796109A (en) * | 2019-11-05 | 2020-02-14 | 哈尔滨理工大学 | Driver distraction behavior identification method based on model fusion |
| CN111178371B (en) * | 2019-12-17 | 2023-12-01 | 深圳市优必选科技股份有限公司 | Target detection method, device and computer storage medium |
| CN111178371A (en) * | 2019-12-17 | 2020-05-19 | 深圳市优必选科技股份有限公司 | Object detection method, device and computer storage medium |
| CN112085056B (en) * | 2020-08-05 | 2023-12-29 | 深圳市优必选科技股份有限公司 | Target detection model generation method, device, equipment and storage medium |
| CN112085056A (en) * | 2020-08-05 | 2020-12-15 | 深圳市优必选科技股份有限公司 | Object detection model generation method, device, device and storage medium |
| CN115131745A (en) * | 2022-06-09 | 2022-09-30 | 合肥讯图信息科技有限公司 | Traffic signal lamp identification method and system under night scene |
| CN119649333A (en) * | 2024-09-18 | 2025-03-18 | 海南省工业研究所 | Traffic sign detection method for complex scenes of autonomous driving based on multi-channel image extraction |
| CN120318782A (en) * | 2025-06-12 | 2025-07-15 | 武汉大学 | Ultra-wide-angle traffic sign detection method and system based on adaptive dynamic pruning |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2016155371A1 (en) | 2016-10-06 |
| JP2017516197A (en) | 2017-06-15 |
| KR20160132331A (en) | 2016-11-17 |
| JP6400117B2 (en) | 2018-10-03 |
| KR101856584B1 (en) | 2018-05-10 |
| CN104700099B (en) | 2017-08-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104700099A (en) | Method and device for recognizing traffic signs | |
| CN113822247B (en) | A method and system for identifying illegal buildings based on aerial images | |
| CN108921083B (en) | Illegal mobile vendor identification method based on deep learning target detection | |
| CN108074244B (en) | Safe city traffic flow statistical method integrating deep learning and background difference method | |
| CN112200143A (en) | Road disease detection method based on candidate area network and machine vision | |
| CN105046235A (en) | Lane line recognition modeling method and apparatus and recognition method and apparatus | |
| CN113255580A (en) | Method and device for identifying sprinkled objects and vehicle sprinkling and leaking | |
| CN109242826B (en) | Mobile equipment end stick-shaped object root counting method and system based on target detection | |
| CN116740652B (en) | Method and system for monitoring rust area expansion based on neural network model | |
| CN112329886A (en) | Double-license plate recognition method, model training method, device, equipment and storage medium | |
| CN112287802A (en) | Face image detection method, system, storage medium and equipment | |
| CN118609069B (en) | Industrial casting monitoring method based on deep learning and graph neural network | |
| CN117152513A (en) | A vehicle boundary positioning method for night scenes | |
| CN117789144A (en) | A cross network lane line detection method and device based on weight fusion | |
| CN116310321B (en) | Target detection method in high-resolution aerial image | |
| CN112949584A (en) | Image screening method and device, electronic equipment and storage medium | |
| CN115049579B (en) | Anti-interference target recognition method based on fusion of multi-band spectral features | |
| CN110163081A (en) | SSD-based real-time regional intrusion detection method, system and storage medium | |
| CN117152665B (en) | Street lamp damage detection method and device and related components | |
| CN117292552B (en) | High-speed road condition analysis system and method based on machine vision | |
| CN114638785B (en) | Method, system, medium, device and terminal for detecting strong light area in image | |
| CN110866435A (en) | Far infrared pedestrian training method with self-similarity gradient oriented histogram | |
| CN117765461A (en) | High-precision people stream evacuation system and method based on image processing | |
| CN110969875B (en) | Method and system for road intersection traffic management | |
| CN115762029A (en) | Alarm method and device based on image detection and electronic equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |