CN115100903A - Highway curve bidirectional early warning system based on YOLOV3 target detection algorithm - Google Patents
Highway curve bidirectional early warning system based on YOLOV3 target detection algorithm Download PDFInfo
- Publication number
- CN115100903A CN115100903A CN202210691153.XA CN202210691153A CN115100903A CN 115100903 A CN115100903 A CN 115100903A CN 202210691153 A CN202210691153 A CN 202210691153A CN 115100903 A CN115100903 A CN 115100903A
- Authority
- CN
- China
- Prior art keywords
- target
- early warning
- curve
- road
- bounding box
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/16—Anti-collision systems
- G08G1/164—Centralised systems, e.g. external to vehicles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/22—Image preprocessing by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/766—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using regression, e.g. by projecting features on hyperplanes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/16—Anti-collision systems
- G08G1/166—Anti-collision systems for active traffic, e.g. moving vehicles, pedestrians, bikes
Landscapes
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Software Systems (AREA)
- Databases & Information Systems (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Image Analysis (AREA)
Abstract
A highway curve bidirectional early warning system based on a YOLOV3 target detection algorithm. The application discloses two-way early warning system who adopts at highway bend highway section includes: the infrared camera with the infrared night vision function is adopted to collect videos of people and vehicles at the road section of the curve, the Yolov3 algorithm is utilized to identify the people and the vehicles at the road section of the curve, advance prompt and early warning are carried out on the people and the vehicles at the other end of the curve by means of the early warning terminal, and a driver is accurately informed of the environments of the vehicles and the roads at the two ends of the curve. The method is suitable for early warning of the road curve section, can effectively avoid traffic safety accidents and improve the traffic safety of the road curve section.
Description
Technical Field
The invention relates to the field of road safety, in particular to a road curve bidirectional early warning system based on a Yolov3 target detection algorithm.
Background
Accidents occur on curved road sections due to a plurality of reasons, and the limitation of the driving sight distance is one of the main reasons. When a vehicle enters a narrow curve road section with a sharp curve, the opposite vehicle and road environments cannot be obtained due to the limitation of factors such as geographical conditions, and serious traffic accidents are easily caused. The former has limited visual field range and is difficult to comprehensively reflect the conditions of people and vehicles on the curve road section, and the former has no adaptability of all-weather work. The latter has great defects in the aspects of early warning dynamics and timeliness, and has limited warning function.
The traffic field aims at a detection method adopting video, and the traditional method of detecting by using pixel values has a plurality of defects in accuracy and classification. The appearance of the target detection technology based on the feature information greatly improves the precision and the class breadth of target detection, and the target detection technology is generally divided into three steps of region selection, feature extraction and classifier classification. Although the early exhaustive target detection framework can realize the improvement of precision to a certain extent, the operation and detection speed is low and time is consumed. Then, a target classification method represented by selective search appeared, which is a great improvement in effect and calculation speed compared with the exhaustive method. With the advent of the R-CNN (regions with CNN features) series, R-CNN uses convolutional neural networks that exhibit superior performance in image classification for the field of target detection. However, the R-CNN framework has many disadvantages, such as that the whole network cannot be end-to-end, a large amount of memory is required to store some features in the intermediate training process, the calculation speed is not ideal, and the like.
Disclosure of Invention
The invention aims to solve the technical problem of providing a bidirectional early warning system for a curved road section based on a Yolov3 target detection algorithm aiming at the defects in the prior art. The YOLO algorithm provides a new idea to convert the problem of target detection into a regression problem, and gives an input image to directly regress the bounding box and classification category of the target at a plurality of positions of the image. The invention provides a detection and early warning system based on a YOLOV3 target detection algorithm and video identification, which can timely and accurately detect and identify complex vehicle and road environments in a road curve section area, provide effective and timely early warning for a driver driving on the curve section, further improve the driving safety of the road curve section and achieve the aim of reducing the traffic accident rate.
The invention provides a road curve two-way early warning system based on a Yolov3 target detection algorithm, which comprises:
step 1, collecting video information of pedestrians and vehicles in a curve road section through cameras with infrared night vision functions arranged on the road sections at two ends and the middle road section of a road curve;
step 2, carrying out feature extraction on video information through a YOLOV3 target detection algorithm; predicting the image position and the class probability value by using the feature extraction result, wherein the class probability value is the conditional probability under the confidence of each bounding box;
step 3, judging a target by using the detected target position and the probability value, and forming a target signal;
step 4, transmitting the target signal to an early warning output unit through the input of a signal transmission unit;
and 5, the early warning output unit realizes acousto-optic output of early warning information after acquiring the target signal of the target identification unit, so as to prompt people and vehicles entering the curve to pay attention to traffic safety.
Preferably, in step 1, the video detection cameras all adopt infrared cameras with an infrared night vision function. The method can be used for collecting images in real time under the condition of dim light and outputting collected image signals without being influenced by the intensity of external light. The light-emitting diode can be used under the condition that the light around the road is dim and even the light intensity at night is almost zero, and the requirement under the condition of dim light is met.
Preferably, the convolutional neural network of Yolo in step 2 divides the input picture into S × S grids, and each cell detects a target whose central point falls within the grid; each cell predicts B bounding boxes and the confidence of the bounding boxes: a bounding box is denoted as pr (object), when the bounding box does not contain a target, pr (object) is 0, and when the bounding box contains a target, pr (object) is 1; the confidence may be defined as
The predictor of each bounding box contains 5 elements: (X, Y, W, H, C), where the first 4 characterize the size and position of the bounding box, and the last value is confidence.
Preferably, the image location and the class probability value are predicted in step 3. The category probability values are conditional probabilities at the confidence of the respective bounding boxes. And (4) performing target judgment by using the detected target position and the probability value, and forming a target signal if traffic participants such as pedestrians, vehicles and the like enter the two sides of the curve and the road section in the curve.
Preferably, the target signal is transmitted to the early warning output unit through the signal transmission unit input in the step 4. The signal transmission unit is used for outputting a corresponding control instruction according to the output result of the target identification unit so as to control whether the early warning output unit outputs specific image information or not.
Preferably, in the step 5, the early warning output unit realizes acousto-optic output of the early warning information after acquiring the target signal of the target identification unit.
The invention selects a famous target detection framework Yolov3 as a tool for target detection, so that the system can normally work in a complex environment.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention, as claimed.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and together with the description, serve to explain the principles of the invention.
Fig. 1 is a technical architecture diagram of a bidirectional early warning system for road curve provided in the present application;
FIG. 2 is a technical flowchart of a bidirectional early warning system for road curves provided by the present application;
fig. 3 is a schematic deployment diagram of a bidirectional early warning system for road curve provided in the present application.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and first, the terms designed by the present invention are explained as follows:
target detection: all objects of interest (objects) in the image are found and their classification and location are determined.
The selective search method comprises the following steps: and extracting the candidate frame by adopting a subregion combination method.
A convolutional neural network: one class of feed-forward neural networks, which include convolution calculations and have a deep structure, is one of the algorithms that represent deep learning.
The invention comprises a target detection unit, a target identification unit, a signal transmission unit and an early warning output unit. The method comprises the steps that cameras are arranged in a curve and on road sections on two sides of the curve to serve as target detection equipment, display early warning equipment is arranged on two sides of the curve, and target recognition equipment is arranged in the curve or on two sides of the curve nearby. The camera is connected with the target recognition device, the target recognition device is connected with the camera and the signal transmission unit, and the signal transmission unit is connected with the target recognition device and the display early warning device.
In the first part, the target detection unit is used for acquiring video images of sections at two ends of a road curve and a middle section of the road curve and inputting the detected video images into the target identification unit. The video detection cameras all adopt infrared cameras with infrared night vision functions. The method can be used for acquiring images in real time under the condition of dim light and outputting acquired image signals without being influenced by the intensity of external light. The light-emitting diode can be used under the condition that the light around the road is dim and even the light intensity at night is almost zero, and the requirement under the condition of dim light is met.
In the second part, after the target identification unit acquires the video image input by the target detection unit, feature extraction is carried out based on a target detection algorithm YOLOV 3. The method comprises the following specific steps:
step 1, performing feature extraction on an input image through a feature extraction network.
The method comprises the following steps of detecting a target by using a YOLO learning framework, wherein a target detection result mainly comprises the following information: the coordinates of the center of the bounding box that encloses the target, the width and height of the bounding box, and the type of bounding box. The YOLO learning framework adopts the high level of YOLOV3 in speed and precision, and the practicability of target detection is greatly improved.
The adopted target detection technology YOLO (YoulokonlyOne) is a target detection frame proposed by Joseph Redmon and the like in 2015, and by changing the traditional frame structure, the first version of YOLO greatly improves the speed of target detection, but does not draw attention due to low accuracy; in 2016, Joseph Redmon et al further proposed YOLOv2, made a partial improvement on YOLO, and while maintaining the speed advantage of YOLO, greatly improved the accuracy of detection, and started to receive extensive attention of researchers; in 2018, joseph redmon et al proposed YOLOv3 again, which made some significant improvements in YOLO, with significant improvements in accuracy while maintaining speed. YOLOv3 processed 512 x 512 resolution pictures on NvidiaGeForce960M GPU on a notebook, up to 10 to 15 frames per second, and could correctly detect most of the targets in the pictures.
The input picture obtains a feature map of 3 scales through the backbone network (from top to bottom: eat1->(256 *52*52),feat2->(512*26*26),feat3->(1024 × 13)), detection was performed on 3 scales, respectively. After 5-layer convolution (Conv2D Block) is carried out on the 3 feature graphs, the three feature graphs respectively enter different branches, one branch is subjected to convolution and upsampling, the obtained feature graph and the feature graph of the upper layer are subjected to channel merging (Concat), and the other branch directly outputs a prediction result through two-layer convolution. The last convolution layer is a 1 x 1 convolution with convolution kernel size (B x (5+ C)). 1 x 1, B represents the number of predicted bounding boxes for a mesh, C represents the probability of C classes, and 5 represents 4 coordinate values (t) and (t) represents the probability of C classes x ，t y ，t w ，t h ) And 1 object confidence. For the coco dataset, C-80 and B-3. The sizes of the final 3-scale measurements were 255 x 52, 255 x 26 and 255 x 52, respectively. One pixel on the feature map corresponds to one mesh in the original image, and each scale defines 3 kinds of anchor frames, that is, each mesh has 3 prediction frames, and each prediction frame has (5+ C) attributes. The nets tested on 3 scales, so the entire net tested a total of 13 × 3+26 × 3+52 × 3 — 10647 bounding boxes.
YOLOv3 determines the boundary box by size clustering along with the method of YOLO9000 predictionAnchor box, predicting bounding boxes of 3 different resolution sizes for each grid point in the 13 × 13 feature map, the net predicting 5 values for each bounding box: t is t x ，t y ，t w ，t h ，t 0 Where the first four are offset values of coordinates, t 0 Is the confidence, the edge distance of the grid from the upper left corner of the image is (c) x ,c y )，p w And p h If the width and height of the prior frame corresponding to the grid are the width and height of the prior frame, the boundary frame prediction is obtained: b x ，b y ，b w ，b h Is the predicted center coordinate x, y of the bounding box, the width b of the bounding box w And a height b h ；
b x ＝σ(t x )+c x
b y ＝σ(t y )+c y
b w ＝p w *e tw
b h ＝p h *e th 。
In the invention, a Yolo convolution neural network is used for dividing an input picture into S multiplied by S grids, and each unit grid detects a target with a central point falling in the grid; each cell predicts B bounding boxes and the confidence of the bounding boxes: a bounding box is denoted as pr (object), when the bounding box does not contain a target, pr (object) is 0, and when the bounding box contains a target, pr (object) is 1; the confidence may be defined as
The predictor of each bounding box contains 5 elements: (X, Y, W, H, C), where the first 4 characterize the size and position of the bounding box, and the last value is the confidence.
And 2, predicting the image position and the category probability value. The category probability values are conditional probabilities at the confidence of the respective bounding boxes.
And 3, performing target judgment by using the detected target position and the probability value, and forming a target signal if traffic participants such as pedestrians, vehicles and the like enter the two sides of the curve and the road section in the curve. The formed target signal is input to the early warning output unit through the signal transmission unit.
And the signal transmission unit is used for outputting a corresponding control instruction according to the output result of the target identification unit so as to control whether the early warning output unit outputs specific image information or not.
And in the fourth step, the early warning output unit realizes acousto-optic output of early warning information after acquiring the target signal of the target identification unit, so as to prompt people and vehicles entering the curve to pay attention to traffic safety.
Claims (6)
1. A two-way early warning system that adopts at highway bend highway section, its characterized in that includes:
step 1, collecting video information of pedestrians and vehicles in a curve road section through cameras with infrared night vision functions arranged on the road sections at two ends and the middle road section of a road curve;
step 2, carrying out feature extraction on video information through a YOLOV3 target detection algorithm; predicting the image position and the category probability value by using the feature extraction result, wherein the category probability value is the conditional probability under the confidence coefficient of each bounding box;
step 3, judging a target by using the detected target position and the probability value, and forming a target signal;
step 4, transmitting the target signal to an early warning output unit through the input of the signal transmission unit;
and 5, the early warning output unit realizes acousto-optic output of early warning information after acquiring the target signal of the target identification unit, so as to prompt people and vehicles entering the curve to pay attention to traffic safety.
2. The method as claimed in claim 1, wherein in step 1, the video detection cameras all use infrared cameras with infrared night vision function.
3. The method according to claim 1, wherein in step 2, the convolutional neural network of Yolo divides the input picture into S × S grids, each cell detects the target whose center point falls within the grid; predicting B bounding boxes and edges per cellConfidence of bounding box: a bounding box is denoted as pr (object), when the bounding box does not contain a target, pr (object) is 0, and when the bounding box contains a target, pr (object) is 1; the confidence may be defined as
The predictor of each bounding box contains 5 elements: (X, Y, W, H, C), where the first 4 characterize the size and position of the bounding box, and the last value is the confidence.
4. The method of claim 1, wherein in step 3, the target is determined by using the detected target position and probability value, and if there is a pedestrian or vehicle entering the road on both sides of the curve or inside the curve, a target signal is formed.
5. The method according to claim 1, wherein in step 4, the signal transmission unit is used for outputting a corresponding control instruction according to the target signal, so as to control whether the early warning output unit outputs the specific image information.
6. The method according to claim 1, wherein in step 5, the early warning output unit realizes acousto-optic output of the early warning information after acquiring the target signal of the target recognition unit.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210691153.XA CN115100903A (en) | 2022-06-17 | 2022-06-17 | Highway curve bidirectional early warning system based on YOLOV3 target detection algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210691153.XA CN115100903A (en) | 2022-06-17 | 2022-06-17 | Highway curve bidirectional early warning system based on YOLOV3 target detection algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115100903A true CN115100903A (en) | 2022-09-23 |
Family
ID=83291325
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210691153.XA Pending CN115100903A (en) | 2022-06-17 | 2022-06-17 | Highway curve bidirectional early warning system based on YOLOV3 target detection algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115100903A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118865742A (en) * | 2024-09-25 | 2024-10-29 | 宁波市威尔信息科技有限公司 | An intelligent traffic feedback system based on Internet of Things technology |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109334563A (en) * | 2018-08-31 | 2019-02-15 | 江苏大学 | An anti-collision warning method based on pedestrians and cyclists in front of the road |
| CN109508710A (en) * | 2018-10-23 | 2019-03-22 | 东华大学 | Based on the unmanned vehicle night-environment cognitive method for improving YOLOv3 network |
| JP2021114059A (en) * | 2020-01-17 | 2021-08-05 | アイシン・エィ・ダブリュ株式会社 | Risk determination system and risk determination program |
| CN113554872A (en) * | 2021-07-19 | 2021-10-26 | 昭通亮风台信息科技有限公司 | Detection early warning method and system for traffic intersection and curve |
| US20210362738A1 (en) * | 2020-05-19 | 2021-11-25 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Information processing method for vehicle driving on curve, electronic device and storage medium |
-
2022
- 2022-06-17 CN CN202210691153.XA patent/CN115100903A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109334563A (en) * | 2018-08-31 | 2019-02-15 | 江苏大学 | An anti-collision warning method based on pedestrians and cyclists in front of the road |
| CN109508710A (en) * | 2018-10-23 | 2019-03-22 | 东华大学 | Based on the unmanned vehicle night-environment cognitive method for improving YOLOv3 network |
| JP2021114059A (en) * | 2020-01-17 | 2021-08-05 | アイシン・エィ・ダブリュ株式会社 | Risk determination system and risk determination program |
| US20210362738A1 (en) * | 2020-05-19 | 2021-11-25 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Information processing method for vehicle driving on curve, electronic device and storage medium |
| CN113554872A (en) * | 2021-07-19 | 2021-10-26 | 昭通亮风台信息科技有限公司 | Detection early warning method and system for traffic intersection and curve |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118865742A (en) * | 2024-09-25 | 2024-10-29 | 宁波市威尔信息科技有限公司 | An intelligent traffic feedback system based on Internet of Things technology |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US9384401B2 (en) | Method for fog detection | |
| CN109740463A (en) | A target detection method in vehicle environment | |
| US20180124319A1 (en) | Method and apparatus for real-time traffic information provision | |
| CN107590470B (en) | Lane line detection method and device | |
| CN106971185B (en) | License plate positioning method and device based on full convolution network | |
| Kang et al. | Application of one-stage instance segmentation with weather conditions in surveillance cameras at construction sites | |
| CN111898491B (en) | Identification method and device for reverse driving of vehicle and electronic equipment | |
| CN111626170B (en) | Image recognition method for railway side slope falling stone intrusion detection | |
| CN102867417A (en) | Taxi anti-forgery system and taxi anti-forgery method | |
| WO2006115676A2 (en) | Video ghost detection by outline | |
| CN111967396A (en) | Processing method, device and equipment for obstacle detection and storage medium | |
| CN104766071A (en) | Rapid traffic light detection algorithm applied to pilotless automobile | |
| CN115601717B (en) | Deep learning-based traffic offence behavior classification detection method and SoC chip | |
| CN115376082B (en) | Lane line detection method integrating traditional feature extraction and deep neural network | |
| CN117789077A (en) | Method for predicting people and vehicles for video structuring in general scene | |
| CN115880658A (en) | Automobile lane departure early warning method and system under night scene | |
| CN115100903A (en) | Highway curve bidirectional early warning system based on YOLOV3 target detection algorithm | |
| JP4577153B2 (en) | Environment recognition device | |
| TWI743637B (en) | Traffic light recognition system and method thereof | |
| CN110688876A (en) | Lane line detection method and device based on vision | |
| CN116630904B (en) | Small target vehicle detection method integrating non-adjacent skip connections and multi-scale residual structure | |
| CN117437808A (en) | A method for over-the-horizon sensing and early warning in blind spots in underground parking lots | |
| CN202887450U (en) | Taxi anti-fake system | |
| CN114155518B (en) | Highway light shield inclination recognition method based on depth semantic segmentation network and image correction | |
| KR102178202B1 (en) | Method and apparatus for detecting traffic light |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |