CN102768801B - Method for detecting motor vehicle green light follow-up traffic violation based on video - Google Patents

Abstract

The invention belongs to the technical field of intelligent traffic management, and in particular relates to a video-based detection method for motor vehicle green light follow-up violations. The steps of the invention are as follows: under the green light phase, when the road ahead is seriously congested, the traffic jam warning sign displays warning information, prompting straight vehicles to prohibit entering the intersection; for motor vehicles that continue to move forward when the warning sign is on, it is listed as a suspected vehicle. Violation, and start track tracking; continue to track suspected violating vehicles, and obtain their complete trajectory at the intersection; the red light phase starts, if the suspected violating vehicles are still stranded in the central no-stop zone, the violation event is completed, and the analysis obtains violation evidence photos and videos. The method has high precision and alertness, and provides a powerful solution to the problem of urban traffic congestion, especially green light follow-up violations.

Description

Video-based detection method for motor vehicle green light follow-up violations

technical field

The invention belongs to the technical field of intelligent traffic management, and in particular relates to a video-based detection method for motor vehicle green light follow-up violations.

Background technique

Green light follow-up is a kind of illegal driving behavior of motor vehicles. Article 53 of the "Regulations for the Implementation of the Road Traffic Safety Law" has specific provisions on this illegal behavior: When motor vehicles encounter traffic jams at the intersection ahead, they should stop at the intersection one by one. Do not wait outside and do not enter the intersection. Its process and impact can refer to Figure 2. The situation in the figure is that when the traffic light is green, the road ahead is already seriously congested. At this time, several vehicles with oblique stripes still enter the intersection. These vehicles will not be able to leave due to congestion, causing traffic lights to stay in the center of the intersection when the traffic lights change, so that vehicles in the crossing direction cannot pass normally, and traffic jams are artificially generated. The correct thing to do in this situation is to wait at the stop line, as indicated by a square-striped vehicle.

This violation has always been a persistent problem of urban traffic congestion. Because once violations are formed, vehicles in the crossing direction may not be able to pass smoothly. In severe cases, the traffic order of the entire intersection will be paralyzed, and traffic police must be relied on to restore order. At present, a common solution used by traffic police in various places is to send traffic police to be on duty at fixed time periods to prevent congestion at intersections with relatively large traffic volumes. This method is a passive solution. On the one hand, it needs to consume a lot of manpower and material resources. On the other hand, since no punishment is imposed, the public's awareness of green light follow-up violations will always remain at a low level. In the long run , is not conducive to the improvement of urban congestion problems.

Since green light follow-up violations can be avoided artificially and have a very fixed pattern, it is feasible to impose penalties on this behavior. There are three broadly speaking ways of judging and punishing: artificial methods, traditional electronic police solutions and video-based electronic police solutions. The following are analyzed separately:

1. Artificial method. At present, in some cities in China, legislation has been carried out on green light follow-up violations. The traffic police have already received preliminary results in punishing illegal vehicles at intersections. However, there are still many disadvantages, such as the inconsistency in the strictness of law enforcement, the failure to punish vehicles for escape in time, and the inability to conduct effective on-site evidence collection. For this type of problem, the electronic police violation automatic recording system can be effectively solved.

2. The traditional electronic police solution. The traditional red light automatic recording system mainly adopts the principle of burying a ground sense coil underground, and when a vehicle passes by the coil, a trigger signal is used to capture the image. In the green light follow-up violation, the two major problems that need to be solved are estimating the state of road congestion and obtaining the trajectory of the vehicle. If the ground induction coil technology wants to solve this violation, it needs to bury a large number of coils in multiple locations underground, design a complicated process, and may rely on other equipment such as radar, and in the end it is difficult to guarantee an acceptable accuracy rate, whether it is It is extremely unfeasible in terms of construction and maintenance costs, construction period, and operational effects.

3. Electronic police solution based on video. In recent years, the electronic police system based on video analysis has gradually become a trend. Compared with the ground induction coil, it has the advantages of accurate capture, low deployment cost, stability and easy maintenance. Chinese patent announcement number CN101419754B has just adopted this type of technology to detect and record running a red light.

The invention also belongs to the electronic police solution based on video analysis. The innovations are: 1) A solution that can detect green light and follow up violations is invented. The automatic detection and recording of violations is the first in China. 2) Congestion status and vehicle trajectory information are used, which can be obtained through video analysis and processing, and other video-based solutions have not yet used such methods.

The violation recording method described in the present invention complies with the national standard GA/T 832-2009, <<Technical Specifications for Image Evidence Collection of Illegal Behaviors in Road Traffic Safety>>, for illegal code 1228 (not waiting in sequence when there is traffic jam at the intersection) . Simultaneously, the present invention also uses the traffic jam warning sign, which is used as auxiliary warning information to prompt the driver of the vehicle to forbid to move forward when congested. This situation is in compliance with the violation code 1231 (Motor Vehicle Violation of Warning Sign Instructions). Therefore, the green light follow-up can be punished according to these two violations.

Contents of the invention

The purpose of the present invention is to propose a simple and effective method for detecting motor vehicle green light follow-up violations.

The motor vehicle green light follow-up violation detection method proposed by the invention uses video to detect and automatically records the vehicle that generates the green light follow-up behavior. The specific steps are:

1. The data acquisition module collects the scene information of the traffic crossing, and specifically uses the camera to continuously shoot the scene to obtain a real-time high-definition video stream; the image of the high-definition video stream should cover the panorama of the crossing, including traffic lights, ground traffic signs, and traffic jam warnings Brand, vehicle characteristics and other information.

2. The video stream is analyzed by the traffic light recognition module, and the current traffic light status information is obtained by using the color and texture features of the traffic light area in the video stream;

3. The video stream is analyzed by the congestion state identification module, and the current road congestion state is identified by using shape and texture information, vehicle feature information, and road flow information; the road congestion state includes two states: congestion and smooth flow. According to the current congestion state, control the traffic jam warning sign to turn on and off.

Congestion state means that due to traffic congestion, there are a large number of vehicles in front that cannot move forward, and the end of the traffic flow composed of these stopped vehicles has reached the opposite zebra crossing. In this state, once subsequent vehicles enter the intersection, they may be stuck in the middle of the intersection and unable to move forward.

4. The vehicle trajectory recognition module analyzes the video stream, mainly using vehicle shape and texture information, and video stream context information to detect vehicles passing through the intersection and identify where the vehicle is currently driving, that is, to obtain vehicle trajectory data . Specifically, the driving trajectory data mainly refers to a set of position coordinates of the vehicle target on the continuous images of the video stream.

5. Combining the status information of traffic lights and congestion status information, the congestion warning sign control module judges and controls whether to display prompt information on the congestion prohibition warning sign.

The significance of the traffic jam warning sign is that in the case of severe congestion, a prompt message is displayed to warn vehicles about to enter the intersection not to move forward. After the warning sign lights up, after a fixed delay time (the delay time is adjustable to facilitate the driver's response), all vehicles that are still entering the intersection will be regarded as suspected illegal vehicles.

6. Combining the current traffic signal status information, congestion status information and vehicle trajectory data, the comprehensive processing module conducts comprehensive processing, specifically including: detecting whether the vehicle has a green light follow-up violation, and for vehicles that have indeed violated the rules, analyze and obtain forensic images And vehicle trajectory video, as a forensic record. Figure 4 is the violation processing flow, the steps are as follows,

(a) Green light phase, when the congestion state is congestion. For a vehicle that crosses the stop line and is about to enter the intersection, set it as a suspected violation vehicle, trigger the violation judgment process for the vehicle, and start tracking the vehicle's trajectory.

(b) Green light phase, when the congestion state is congestion. If a vehicle suspected of violating the rules enters the intersection, the process of judging the vehicle's violations will be maintained, and it will be tracked continuously to obtain the driving trajectory.

(c) After the green light phase changes to the red light phase, if the suspected violating vehicle still stays at the intersection, the violation event is completed. From the analysis of the video stream, the forensic image that can be seen that the vehicle enters the intersection and stays at the intersection is obtained, and the video containing the illegal driving process of the vehicle is stored.

(d) other circumstances. Under the green light phase, the congestion state changes to no congestion. The current suspected violation vehicle does not constitute a violation, and the violation judgment process ends.

(e) For other situations that do not meet the procedures (a), (b), and (c), the suspected violation vehicles do not constitute actual violations, and the violation judgment process ends.

The module structure of the video analysis used in this method can be seen in FIG. 3 .

Description of drawings

Fig. 1 is a schematic diagram of the scene and steps of the present invention. Among them, 1 is a high-definition camera, and 2 is a traffic jam warning sign.

Figure 2 is a schematic diagram of green light follow-up behavior.

Fig. 3 is a block diagram of video analysis in the method of the present invention.

Fig. 4 is a flow chart of judging violations of the present invention.

Detailed ways

The specific implementation manner of the present invention will be described below in conjunction with the accompanying drawings.

The application scenario of this method is shown in Figure 1. A high-definition camera is deployed on the side of the intersection. The camera has a resolution of more than 2 million, and the field of view can cover 2 to 3 lanes. The traffic jam warning sign should be installed near the opposite traffic light, so that passing vehicles can clearly see the prompt content of the warning sign, and the warning sign should also appear within the range of the camera screen. Preferably, an LED display is used as a warning sign.

The video processing module structure of this method is formed according to the description in Figure 3, the input is the original video data collected by the high-definition camera, and the output is the warning sign control signal and the green light follow-up violation evidence collection record. The comprehensive processing module will store evidence collection records and upload them to the traffic police law enforcement department through the network. The input source of this method only needs a camera, and uses pure video processing to detect and automatically record violations, without access to traffic light signals or ground sense coils. Compared with other electronic police automatic recording schemes, this has a small deployment cost and construction period. Through the quick configuration of software-side parameters, this solution can be flexibly applied to various intersections where green light follow-up behaviors may occur. It has strong versatility and is suitable for large-scale promotion.

Figure 3 includes a data collection module, a traffic light detection and recognition module, a congestion state recognition module, a vehicle trajectory recognition module, a warning sign control module and a comprehensive processing module. in,

The data acquisition module decodes the high-definition video data. For example, convert h264 format data into frame-by-frame RGB images, and output frame number, timestamp and other related parameters. Subsequent video analysis is carried out in RGB images.

Traffic light detection and recognition module. Its input is a single frame image, which contains traffic lights at intersections. First, the module will use the characteristics of traffic lights such as color and texture to train the traffic light model offline. During system operation, the module uses the trained model to detect traffic light locations and identify current signals. The output information is the current phase state, red or green.

Congestion state identification module. Its input is a single frame of image that contains vehicles on the oncoming road. The output is the state information of road congestion. The judgment condition is

in is the ingress traffic flow (the number of passing vehicles per second), is the export traffic flow, is the number of lanes of the road, is the congestion judgment threshold, generally set to 1.5.

Vehicle trajectory recognition module. Its input is a video stream, and a single frame image of the video stream contains the entire scene of the intersection on this side. The output is vehicle trajectory information, which contains the actual coordinates of the vehicle on the road. This module uses computer vision-based algorithms, which contain two important components, vehicle detection and vehicle tracking. Among them, vehicle detection uses vehicle shape, texture and other characteristics to obtain the position coordinates of the vehicle entering the monitoring range for the first time; vehicle tracking uses shape and video stream context information to constantly judge and obtain the current vehicle position coordinates in real time. The results of the two together constitute the vehicle trajectory information.

Warning sign control module. The input is the switch notification message obtained by the analysis module, and the output is the warning sign control signal. When the road is congested, control the warning sign to open.

Comprehensive processing module. The input is the current traffic light phase state, congestion state, vehicle trajectory, and image sequence.

The output is to judge whether a violation is achieved, and if so, save the evidence of violation. The internal processing of the module is carried out according to the process described in Figure 4.

A typical green light follow-up violation record process carried out by the present invention is described in FIG. 4 . First, refer to A. In the green light phase, when the congestion state is congestion, the congestion warning sign will be on. B. After the warning sign lights up, if there is a vehicle that ignores the warning sign prompts and still drives across the zebra crossing, it will be listed as a suspected violation vehicle. Generally, between A and B, there will be a delay as the reaction time for the driver to see the plate light up and stop driving. Afterwards, before C, if the warning sign goes off (that is, the congestion state turns to unblocked), the vehicle does not constitute a violation, and the process is terminated. C. If a vehicle suspected of violating the rules enters the central no-stop area of the intersection, keep track of the vehicle and determine the violation process. Afterwards, before D, if the congestion is lifted, it does not constitute a violation, and the process is terminated. Finally, in D, the traffic light changes from green light to red light, and after the red light is on for a period of time, if the suspected illegal vehicle fails to drive out and stays in the middle of the intersection, it will be determined as a green light follow-up violation, and the analysis can obtain its Forensic images and videos of violations are used as law enforcement evidence.

This kind of violation recording method adopts the recording method of pictures and videos. The forensic pictures can clearly see that the vehicles pass the zebra crossing, stay at the intersection at the red light, and can clearly distinguish the license plate. The video contains the complete violation process, which has a corresponding relationship with the forensics pictures, and serves as auxiliary evidence. On the basis of meeting the national standard GA/T832-2009 illegal evidence collection mode 1, this method also adds video auxiliary evidence. The forensic data obtained according to this method has extremely high validity, and it is not easy to cause disagreements in law enforcement.

The above descriptions are preferred embodiments of the present invention, and are not intended to limit the implementation scope of the present invention. All approximate changes and modifications made according to the present invention belong to the protection scope of the claims of the present invention.

Claims (3)

1. A method for detecting motor vehicle green light follow-up violation based on video is characterized by comprising the following specific steps:

(1) acquiring scene information of a traffic crossing by a data acquisition module, and continuously shooting the scene by a camera to obtain a real-time high-definition video stream;

(2) analyzing the video stream by the traffic light identification module, and acquiring the current traffic signal light state information by utilizing the color and texture characteristics of the traffic light area in the video stream;

(3) analyzing the video stream by a congestion state identification module, and identifying and obtaining the current road congestion state by utilizing shape and texture information, vehicle characteristic information and road flow information; the road congestion state comprises a congestion state and a smooth state; controlling the congestion driving prohibition warning board to light up and light down according to the current congestion state;

the congestion state refers to a traffic flow formed by a plurality of stopped vehicles which cannot drive forwards due to traffic congestion, and the tail ends of the traffic flow reach opposite zebra crossings;

(4) analyzing the video stream by a vehicle track recognition module, mainly utilizing vehicle shape and texture information and video stream context information, detecting vehicles passing through an intersection, and recognizing where the current vehicle runs to obtain vehicle running track data, wherein the running track data comprises a group of position coordinates of a vehicle target on continuous images of the video stream;

(5) the congestion warning board control module judges and controls whether the congestion no-go warning board displays the prompt information or not by combining the traffic signal lamp state information and the congestion state information;

when the congestion is serious, displaying prompt information to warn vehicles about to enter the intersection not to move forward; when the congestion warning board is lighted up and a fixed delay time is passed, all vehicles still driving into the intersection are regarded as suspected violation vehicles;

(6) and combining the current traffic signal lamp state information, the congestion state information and the vehicle running track data, and carrying out comprehensive processing by a comprehensive processing module, wherein the comprehensive processing module specifically comprises: detecting whether the vehicle is in green light follow-up violation, analyzing and obtaining a evidence obtaining image and a vehicle track video for the vehicle which really has the violation as an evidence obtaining record, and specifically comprising the following steps,

(a) when the congestion state is congestion, setting the vehicles which cross the stop line and are about to enter the intersection as suspected violation vehicles, triggering a violation judgment process aiming at the vehicles, and starting to track the vehicles;

(b) if the suspected vehicle violating the regulations enters the intersection, keeping a violation judgment process for the vehicle, and continuously tracking the vehicle violating the regulations to obtain a vehicle running track;

(c) after the green light phase is changed into the red light phase, if the suspected violation vehicle is still remained at the intersection, the violation event is achieved; then, obtaining evidence-obtaining images which can show that the vehicle enters the intersection and stays at the intersection from the video stream by analysis, and storing the video containing the vehicle illegal driving process;

(d) under the green light phase, the congestion state is changed into non-congestion, the current suspected violation vehicle does not form violation, and the violation judgment process is ended;

(e) and (c) for other conditions which do not meet the flow of (a), (b) and (c), the suspected violation vehicles do not form actual violation, and the violation judgment process is ended.

2. The detection method according to claim 1, wherein the image of the high-definition video stream in step (1) covers a crossing panorama, including traffic signal lights, ground traffic signs, congestion warning boards, and vehicle body feature information.

3. The detection method according to claim 2, wherein the congestion prohibition warning board is placed in a visual field visible to both the driver and the camera.