CN105160340A - Vehicle brand identification system and method - Google Patents

Abstract

The invention relates to a vehicle brand recognition system and method, the system includes a monitoring video acquisition module, a vehicle detection module, a vehicle logo segmentation module, a vehicle logo feature extraction module and a vehicle brand recognition module; the monitoring video acquisition module obtains the traffic monitoring checkpoint Vehicle monitoring video; the vehicle detection module detects the vehicle on the vehicle monitoring video, obtains the position of all vehicles in the vehicle monitoring video, and intercepts the vehicle picture according to the position information of the vehicle; the vehicle logo segmentation module detects the car logo from the vehicle picture, and obtains the vehicle According to the location of the car logo, the car logo image is segmented; the car logo feature extraction module extracts the car logo features from the car logo image; the vehicle brand recognition module classifies the extracted car logo feature information, and obtains the vehicle brand according to the feature classification results. . Compared with the prior art, the invention has low cost, wide application range, fast recognition speed and high accuracy.

Description

A vehicle brand recognition system and method

technical field

The invention relates to the technical fields of pattern recognition and intelligent traffic information, in particular to a vehicle brand recognition system and method.

Background technique

In recent years, with the continuous progress of society, intelligent transportation systems have developed rapidly. At the same time, the development of computer vision and pattern recognition technology has provided opportunities for more effective applications of intelligent transportation systems. Computer vision is the use of computers to simulate human visual functions, extract information from images of objective things, process and understand them, and finally use them for actual detection, measurement and recognition.

In intelligent transportation system, vehicle brand recognition is an important part. Vehicle brand recognition is based on the uniqueness of the appearance and logo of different brands of vehicles. It has broad application prospects in urban traffic monitoring, emergency command, accident detection, intelligent path guidance and other fields.

In the existing vehicle brand recognition scheme, it mainly consists of three parts: firstly, the front image block is intercepted from the vehicle picture; secondly, the characteristic information of the front image block is extracted; finally, the characteristic information is input into the classifier to obtain the vehicle brand recognition result .

In the first part, the car front picture is intercepted from the vehicle picture. Although the car front picture contains certain brand information of the vehicle, it also contains a lot of redundant information and interference information, such as the interference of the front cooling net and the license plate. The redundant information and interference information brought by intercepting the front image will bring many difficulties to the subsequent feature extraction and classification.

In the second part, the feature information of the front image block is extracted. The existing technology mainly extracts the surface features of the image, including histogram of gradient orientation (HoG), linear binarization model (LBP), etc. It reflects the shape of the image well and is robust to changes in illumination and angles, but this type of feature information is not rich enough to meet the classification requirements of a large number of vehicle brands.

With the advancement of existing traffic monitoring technology, high-definition cameras are widely used. In particular, high-definition cameras can obtain clear pictures of car logos from moving vehicles, which also provides favorable conditions for technological breakthroughs in vehicle brand recognition based on surveillance video. Aiming at the deficiencies in the prior art, the present invention proposes a vehicle brand recognition method and system based on the analysis of vehicle logo pictures. The method accurately obtains the vehicle logo pictures from the traffic monitoring video, and extracts the deep features of the car logo pictures. Information, classify the obtained feature information and finally identify the vehicle brand.

Contents of the invention

The technical problem to be solved by the present invention is to provide a vehicle brand recognition system and method with low cost, wide application range, fast recognition speed and high accuracy.

The technical solution of the present invention to solve the above-mentioned technical problems is as follows: a vehicle brand recognition system, including a monitoring video acquisition module, a vehicle detection module, a vehicle logo segmentation module, a vehicle logo feature extraction module and a vehicle brand recognition module; the monitoring video acquisition module , vehicle detection module, vehicle logo segmentation module, vehicle logo feature extraction module and vehicle brand recognition module are connected in sequence;

The monitoring video acquisition module is used to acquire the vehicle monitoring video at the traffic monitoring checkpoint;

The vehicle detection module is used to detect the vehicle on the vehicle monitoring video, obtain the positions of all vehicles in the vehicle monitoring video, and intercept the vehicle picture according to the position information of the vehicle;

The car logo segmentation module is used to detect the car logo from the vehicle picture, obtain the position of the car logo, and segment the car logo picture according to the position of the car logo;

The vehicle logo feature extraction module is used to extract the car logo feature from the car logo picture;

The vehicle brand identification module is used to classify the extracted vehicle logo feature information, and obtain the vehicle brand according to the feature classification results.

The beneficial effects of the present invention are: using monitoring video as the input information of the vehicle brand recognition system can reduce equipment investment and effectively reduce costs, and the existing traffic checkpoint monitoring network has a wide coverage and can obtain a wide range of vehicle monitoring video; The detection module extracts vehicle pictures from the traffic checkpoint monitoring video, narrows the search range for vehicle logo detection, and improves the accuracy of vehicle logo detection; the vehicle logo segmentation module directly detects vehicle logos on vehicle pictures and extracts As the only feature image for vehicle brand recognition, pictures can effectively reduce redundant information and interference information, and improve the accuracy of vehicle logo recognition; the vehicle logo feature extraction module can extract the feature information of car logo pictures, and different car logo pictures can get different The characteristic information can use the difference of the characteristic information of the vehicle logo picture to identify the vehicle logo; the vehicle brand recognition module can classify the characteristic information of the car logo picture, and can identify different car logo graphics, because the car logo picture and the vehicle brand There is a one-to-one correspondence, that is, one type of vehicle logo image corresponds to only one type of vehicle brand. Therefore, vehicle brand recognition can be realized by utilizing the correspondence between the vehicle logo image and the vehicle brand.

On the basis of the above technical solutions, the present invention can also be improved as follows.

Further, the surveillance video acquisition module includes more than one camera.

Further, the vehicle detection module uses image pattern recognition technology to detect vehicles and obtain pictures of vehicles.

The beneficial effect of adopting the above further solution is that: using the image pattern recognition technology for vehicle detection improves the efficiency of vehicle recognition.

Further, the vehicle detection module performs vehicle detection on a single frame of the surveillance video, obtains vehicle position information in the single frame, and intercepts a vehicle picture according to the vehicle position information.

The beneficial effect of adopting the above further solution is that the vehicle detection is performed on a single frame of the surveillance video, which can identify both moving vehicles and stationary vehicles, thereby expanding the scope of application of the system.

Further, the vehicle logo segmentation module performs a car logo search near the symmetry axis of the vehicle picture according to the symmetry of the vehicle picture.

The beneficial effect of adopting the above further solution is that the symmetrical search can improve the search efficiency and the recognition speed.

Further, the vehicle logo feature extraction module extracts the car logo feature from the car logo picture through a deep neural network.

The beneficial effect of adopting the above-mentioned further solution is that the deep neural network can improve the accuracy of extracting the features of the vehicle logo image and ensure the accuracy of the vehicle logo recognition.

Further, the vehicle brand recognition module includes a classifier, the classifier classifies the extracted vehicle logo feature information, and outputs the probability value that the vehicle logo picture belongs to each type of vehicle logo.

Further, the vehicle brand recognition module compares each probability value output by the classifier with a set probability threshold, and if each probability value is less than the probability threshold, it is determined that the input is not a vehicle logo picture; if only one probability value is greater than the probability threshold threshold, it is determined that the image is the vehicle logo image corresponding to the maximum probability value; if there are multiple probability values greater than the probability threshold, it is reclassified, and the probability value is compared with the probability threshold.

The beneficial effect of adopting the above-mentioned further solution is that the output of the classifier is the probability value, which can perform secondary discrimination according to the numerical value of each probability, and effectively improve the recognition accuracy of the vehicle logo.

Another technical solution of the present invention to solve the above-mentioned technical problems is as follows: a vehicle brand recognition method, comprising the following steps:

Step S1. Obtain the vehicle monitoring video at the traffic monitoring checkpoint;

Step S2. Carry out vehicle detection on the vehicle monitoring video, obtain the positions of all vehicles in the vehicle monitoring video, and intercept the vehicle picture according to the position information of the vehicle;

Step S3. Detect the vehicle logo from the vehicle image, obtain the position of the vehicle logo, and segment the vehicle logo image according to the position of the vehicle logo;

Step S4. Carry out car logo feature extraction to the car logo picture;

Step S5. Classify the extracted vehicle logo feature information, and obtain the vehicle brand according to the feature classification result.

The beneficial effects of the present invention are: using the monitoring video as the input information of the vehicle brand recognition system can reduce equipment investment, effectively reduce the cost, and expand the scope of use of the system; extract vehicle images from the monitoring video and narrow the search range of the vehicle logo; Car logo detection and extraction of car logo pictures from vehicle images can effectively reduce redundant information and interference information; extract feature information of car logo pictures, different car logo images can get different feature information, and the difference in feature information of car logo pictures can be used Vehicle logo recognition; the feature information of the car logo picture is classified into the car logo, and different car logo pictures can be identified, and the vehicle brand recognition can be realized by using the corresponding relationship between the car logo picture and the vehicle brand.

Further, the specific implementation of the step S5: classify the extracted vehicle logo feature information, compare each probability value of the classification output with the set probability threshold, if each probability value is less than the probability threshold, it is determined that the input is not Vehicle logo pictures; if only one probability value is greater than the probability threshold, it is determined that the image is the vehicle logo picture corresponding to the maximum probability value; if there are multiple probability values greater than the probability threshold, it is reclassified, and the probability value is compared with the probability threshold right.

The beneficial effect of adopting the above-mentioned further solution is that: the classification output is the probability value, and secondary discrimination can be performed according to the numerical value of each probability, thereby effectively improving the recognition accuracy of the vehicle logo.

Description of drawings

Fig. 1 is a block diagram of a vehicle brand recognition system of the present invention;

Fig. 2 is a flowchart of a vehicle brand recognition method of the present invention.

In the accompanying drawings, the list of parts represented by each label is as follows:

1. Monitoring video acquisition module, 2. Vehicle detection module, 3. Vehicle logo segmentation module, 4. Vehicle logo feature extraction module, 5. Vehicle brand recognition module.

Detailed ways

The principles and features of the present invention are described below in conjunction with the accompanying drawings, and the examples given are only used to explain the present invention, and are not intended to limit the scope of the present invention.

Example 1:

As shown in Figure 1, a kind of vehicle brand identification system comprises monitoring video acquisition module 1, vehicle detection module 2, vehicle logo segmentation module 3, vehicle logo feature extraction module 4 and vehicle brand identification module 5; described monitoring video acquisition module 1. The vehicle detection module 2, the vehicle logo segmentation module 3, the vehicle logo feature extraction module 4 and the vehicle brand recognition module 5 are sequentially connected;

The monitoring video acquisition module 1 is used to acquire the vehicle monitoring video at the traffic monitoring checkpoint;

The vehicle detection module 2 is used to detect the vehicle on the vehicle monitoring video, obtain the positions of all vehicles in the vehicle monitoring video, and intercept the vehicle picture according to the position information of the vehicle;

The car logo segmentation module 3 is used to detect the car logo from the vehicle picture, obtain the position of the car logo, and segment the car logo picture according to the position of the car logo;

The vehicle logo feature extraction module 4 is used to extract the car logo feature to the car logo picture;

The vehicle brand identification module 5 is used to classify the extracted vehicle logo feature information, and obtain the vehicle brand according to the feature classification results.

Preferably, the surveillance video acquisition module 1 includes more than one camera.

Preferably, the vehicle detection module 2 uses image pattern recognition technology to detect vehicles and acquire pictures of vehicles.

Preferably, the vehicle detection module 2 performs vehicle detection on a single frame of the surveillance video, obtains vehicle position information in the single frame, and intercepts a vehicle picture according to the vehicle position information.

Preferably, the vehicle logo segmentation module 3 searches for the vehicle logo near the symmetry axis of the vehicle picture according to the symmetry of the vehicle picture.

Preferably, the vehicle logo feature extraction module 4 extracts the car logo feature from the car logo picture through a deep neural network.

Preferably, the vehicle brand identification module 5 includes a classifier, the classifier classifies the extracted vehicle logo feature information, and outputs the probability value that the vehicle logo picture belongs to each type of vehicle logo.

Preferably, the vehicle brand recognition module 5 compares each probability value output by the classifier with a set probability threshold, and if each probability value is less than the probability threshold, it is determined that the input is not a car logo picture; if there is only one probability value If it is greater than the probability threshold, it is determined that the image is the vehicle logo image corresponding to the maximum probability value; if there are multiple probability values greater than the probability threshold, it will be reclassified, and the probability value is compared with the probability threshold.

As shown in Figure 2, a kind of vehicle brand recognition method comprises the following steps:

Step S1. Obtain the vehicle monitoring video at the traffic monitoring checkpoint;

Step S2. Carry out vehicle detection on the vehicle monitoring video, obtain the positions of all vehicles in the vehicle monitoring video, and intercept the vehicle picture according to the position information of the vehicle;

Step S3. Detect the vehicle logo from the vehicle image, obtain the position of the vehicle logo, and segment the vehicle logo image according to the position of the vehicle logo;

Step S4. Carry out car logo feature extraction to the car logo picture;

Step S5. Classify the extracted vehicle logo feature information, and obtain the vehicle brand according to the feature classification result.

Preferably, the specific implementation of step S5: classify the extracted vehicle logo feature information, compare each probability value output by classification with the set probability threshold, if each probability value is less than the probability threshold, then determine the input It is not a car logo picture; if there is only one probability value greater than the probability threshold, it is determined that the image is the car logo picture corresponding to the maximum probability value; if there are multiple probability values greater than the probability threshold, it is reclassified, and the probability value and the probability threshold are compared. Comparison.

Example 2:

A vehicle brand recognition method, comprising the following steps:

Step 1. The traffic checkpoint monitoring video acquisition module is composed of one or more high-definition cameras, which acquires the vehicle video information passing through the traffic monitoring checkpoint, and provides original video information for the entire vehicle brand recognition system;

Step 2. The vehicle detection module detects the vehicle on the traffic checkpoint monitoring video, obtains the positions of all vehicles in each frame of the video screen, and intercepts the vehicle picture according to the position information of the vehicle;

Among them, the vehicle detection uses a sliding window method to obtain image blocks of different positions and sizes in a single frame of video, and extracts the gradient direction histogram of these image blocks as feature information, and inputs the feature information of these image blocks into training. The support vector machine of the support vector machine judges whether the input graphic block is a vehicle image by the support vector machine, thus detects all the vehicles in the single-frame video picture;

Step 3. The vehicle logo segmentation module first uses the symmetry of the vehicle image to obtain the symmetry axis of the vehicle image, and then obtains image blocks of different positions and sizes in the middle of the vehicle image along the symmetry axis; extracts the gradient direction histogram of the image block as feature information , the feature information of these image blocks is input into the trained support vector machine, and the support vector machine judges whether the input graphic block is a car logo image, thereby detecting the car logo in the vehicle image and segmenting the car logo image;

Step 4. Use the deep neural network to extract the deep feature information of the car logo image. The deep neural network is composed of 3 convolutional layers, 2 average pooling layers, and 1 fully connected layer. The output vector of the fully connected layer is the car The depth feature information of the target image;

Step 5. Input the feature information of the car logo image into the softmax classifier, and the output vector of the softmax classifier is the probability value that the car logo image belongs to each brand car logo. Then, compare each probability value output by the classifier with the set probability threshold, if they are all smaller than the probability threshold, it is determined that the input is not a car logo image; if only one probability value is greater than the probability threshold, then it is judged that the input car logo image Belong to the type of vehicle logo corresponding to the maximum probability value; if there are multiple probability values greater than the threshold, use a new classifier for classification, and repeat the above classification steps. Identify the brand to which the vehicle belongs based on the relationship between the logo image and the vehicle brand;

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection of the present invention. within range.

Claims (10)

1. a vehicle brand recognition system, is characterized in that: comprise monitor video acquisition module (1), vehicle detection module (2), car mark segmentation module (3), car mark characteristic extracting module (4) and vehicle brand identification module (5); Described monitor video acquisition module (1), vehicle detection module (2), car mark segmentation module (3), car mark characteristic extracting module (4) are connected successively with vehicle brand identification module (5);

Described monitor video acquisition module (1), for obtaining the vehicle monitoring video of traffic monitoring bayonet socket;

Described vehicle detection module (2), for carrying out vehicle detection to vehicle monitoring video, obtains the position that in vehicle monitoring video, all vehicles occur, and intercepts vehicle pictures according to the positional information of vehicle;

Described car mark segmentation module (3), for inspection vehicle mark from vehicle pictures, obtains car cursor position, and is partitioned into car according to car cursor position and marks on a map sheet;

Described car mark characteristic extracting module (4), carries out the feature extraction of car mark for sheet of marking on a map to car;

Described vehicle brand identification module (5), for classifying to the car mark characteristic information extracted, obtains vehicle brand according to tagsort result.

2. a kind of vehicle brand recognition system according to claim 1, is characterized in that: described monitor video acquisition module (1) comprises the video camera of more than.

3. a kind of vehicle brand recognition system according to claim 1, is characterized in that: described vehicle detection module (2) uses image pattern recognition to carry out vehicle detection, obtains vehicle pictures.

4. a kind of vehicle brand recognition system according to claim 1, it is characterized in that: the single frames picture of described vehicle detection module (2) to monitor video carries out vehicle detection, and the vehicle position information obtained in single frames picture, the positional information according to vehicle intercepts vehicle pictures.

5. a kind of vehicle brand recognition system according to claim 1, is characterized in that: described car mark segmentation module (3), according to the symmetry of vehicle pictures, carries out the search of car mark near vehicle pictures axis of symmetry.

6. a kind of vehicle brand recognition system according to claim 1, is characterized in that: described car mark characteristic extracting module (4) carries out the feature extraction of car mark by deep neural network to car sheet of marking on a map.

7. a kind of vehicle brand recognition system according to any one of claim 1 to 6, it is characterized in that: described vehicle brand identification module (5) comprises sorter, described sorter is classified to the car mark characteristic information extracted, and exports car sheet of marking on a map and belong to all kinds of car target probable value.

8. a kind of vehicle brand recognition system according to claim 7, it is characterized in that: the probability threshold value of each probable value that sorter exports by described vehicle brand identification module (5) and setting is made comparisons, if each probable value is all less than probability threshold value, then what judge input is not that car is marked on a map sheet; If only have a probable value to be greater than probability threshold value, then judge that this image to be marked on a map sheet as the car that most probable value is corresponding; If there is multiple probable value to be greater than probability threshold value, then reclassify, carry out probable value and probability threshold values is compared.

9. a vehicle brand recognition methods, is characterized in that, comprises the following steps:

Step S1. obtains the vehicle monitoring video of traffic monitoring bayonet socket;

Step S2. carries out vehicle detection to vehicle monitoring video, obtains the position that in vehicle monitoring video, all vehicles occur, and intercepts vehicle pictures according to the positional information of vehicle;

Step S3. is inspection vehicle mark from vehicle pictures, obtains car cursor position, and is partitioned into car according to car cursor position and marks on a map sheet;

Step S4. carries out the feature extraction of car mark to car sheet of marking on a map;

Step S5. classifies to the car mark characteristic information extracted, and obtains vehicle brand according to tagsort result.

10. a kind of vehicle brand recognition methods according to claim 9, it is characterized in that: the specific implementation of described step S5: the car mark characteristic information extracted is classified, each probable value export classification and the probability threshold value of setting are made comparisons, if each probable value is all less than probability threshold value, then what judge input is not that car is marked on a map sheet; If only have a probable value to be greater than probability threshold value, then judge that this image to be marked on a map sheet as the car that most probable value is corresponding; If there is multiple probable value to be greater than probability threshold value, then reclassify, carry out probable value and probability threshold values is compared.