Method for discovering non-deceleration of motor vehicle at intersection based on road monitoring video

Abstract

The invention relates to the technical field of image recognition, and particularly discloses a motor vehicle intersection non-deceleration discovery method based on a road monitoring video, and the method comprises the steps: obtaining a road monitoring video of a target intersection; respectively tracking traffic lights in the current direction, motor vehicles in the current direction and motor vehicles in other directions in the road monitoring video of the target intersection so as to correspondingly obtain the traffic light change time in the current direction, the motor vehicle motion trail in the current direction and the motor vehicle motion trails in other directions of the target intersection; wherein the motion trail of the motor vehicle in the current direction comprises stop line passing time of the motor vehicle in the current direction; and according to the traffic light change time in the current direction of the target intersection, the motor vehicle motion trail in the current direction and the motor vehicle motion trails in other directions, judging whether the motor vehicle in the current direction does not decelerate when passing through the target intersection. According to the motor vehicle intersection non-deceleration discovery method based on the road monitoring video provided by the invention, the probability of occurrence of traffic accidents at the intersection can be reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition, and more specifically, to a method for detecting a motor vehicle that does not slow down at an intersection based on a road monitoring video. Background Art

[0002] It is an uncivilized traffic behavior for a motor vehicle to pass through an intersection at high speed when it has the right of way. It can easily lead to serious traffic accidents, causing avoidable accidents or making originally minor accidents more serious.

[0003] Traffic management departments generally set up speed limit signs at intersections. For example, urban intersections are generally limited to 30 kilometers per hour, and speed reduction markings and speed reduction signs are set up at intersections to curb traffic violations such as not slowing down when passing through intersections.

[0004] However, it is difficult to continuously track and monitor the speed of vehicles before and after entering the intersection due to the existing electronic police equipment. It is not easy to collect evidence for the illegal behavior of not slowing down at the intersection. Therefore, there is a general lack of equipment to monitor the speed of vehicles at intersections, resulting in the uncivilized behavior of motor vehicles not slowing down at intersections. The current situation is that it is not only difficult to collect evidence, but also difficult to be discovered. Traffic management departments in various places are not only unable to carry out non-site penalties, but also unable to implement control measures such as reminders and warnings.

[0005] Given that road video surveillance equipment is generally installed at road intersections in various places, detecting motor vehicles not slowing down at intersections based on surveillance video can maximize the detection of illegal behaviors of motor vehicles not slowing down at intersections without increasing investment, and provide support for control measures such as reminders, warnings, and on-site penalties. Therefore, a method for detecting motor vehicles not slowing down at intersections based on road surveillance video is urgently needed.

[0006] However, on the one hand, it is difficult to accurately calculate the vehicle speed in road surveillance videos due to the lack of calibration and perspective. On the other hand, the video surveillance distance is short and the vehicle speed change is not obvious. Directly calculating the vehicle speed through the video to detect illegal behaviors such as not slowing down at intersections is likely to cause false alarms and missed alarms.

[0007] The present invention attempts to indirectly detect the illegal behavior of not slowing down at an intersection. The main basis is that passing through an intersection at high speed is a dangerous driving behavior, which will affect the behavior of other vehicles and pedestrians. For example, vehicles that "rush to the green light" to pass through an intersection at high speed before the green light turns red are likely to affect vehicles turning in other directions, causing vehicles to stop suddenly, change their trajectories, etc. Through these phenomena, it is possible to indirectly detect that motor vehicles do not slow down at intersections. Summary of the invention

[0008] In order to solve the deficiencies in the prior art, the present invention provides a method for detecting a motor vehicle that does not slow down at an intersection based on road monitoring video, which can reduce the probability of traffic accidents at intersections.

[0009] As a first aspect of the present invention, a method for detecting a motor vehicle that does not slow down at an intersection based on a road monitoring video is provided, comprising the following steps: Step S1: Obtain road surveillance video of the target intersection; Step S2: Track the current direction traffic light, the current direction motor vehicle and the other direction motor vehicles in the road monitoring video of the target intersection respectively, so as to obtain the current direction traffic light change time, the current direction motor vehicle movement trajectory and the other direction motor vehicle movement trajectory of the target intersection; wherein the current direction motor vehicle movement trajectory includes the current direction motor vehicle passing the stop line time; Step S3: According to the traffic light change time of the current direction of the target intersection, the movement trajectory of the motor vehicle in the current direction and the movement trajectory of the motor vehicles in other directions, it is determined whether the motor vehicle in the current direction does not slow down when passing the target intersection.

[0010] Furthermore, the step S3 further includes: According to the time when the traffic light in the current direction of the target intersection changes to green or green and the time when the motor vehicle in the current direction passes the stop line, the time interval between the time when the motor vehicle in the current direction passes the stop line and the time when the red light in the current direction turns on is calculated; If the time interval between the time when the motor vehicle in the current direction passes the stop line and the time when the red light in the current direction turns on is less than the standard value, the motion trajectory of motor vehicles in other directions of the target intersection is identified by the video trajectory recognition model; When it is identified that the movement trajectory of the motor vehicle in the other direction is abnormal, it is determined that the motor vehicle in the current direction is suspected of not slowing down when passing the target intersection; The speed of the motor vehicle in the current direction passing through the target intersection is calculated. When the speed of the motor vehicle in the current direction passing through the target intersection is greater than a preset value, it is determined that the motor vehicle in the current direction does not decelerate when passing through the target intersection.

[0011] Further, the step of calculating the speed of the motor vehicle in the current direction passing through the target intersection further includes: Using a vehicle type recognition algorithm to identify the vehicle type of the motor vehicle in the current direction in the road monitoring video of the target intersection; When the vehicle type of the motor vehicle in the current direction is identified, the actual length of the vehicle body in the current direction is obtained according to the existing vehicle type registration information; When the vehicle type of the motor vehicle in the current direction cannot be identified, the vehicle body pixel length of the motor vehicle in the current direction is extracted from the road monitoring video of the target intersection, and the actual vehicle body length of the motor vehicle in the current direction is obtained according to the vehicle body pixel length and the actual vehicle body length of the motor vehicle of the identified vehicle type in the road monitoring video of the target intersection; The speed of the motor vehicle in the current direction passing through the target intersection is calculated based on the actual length of the motor vehicle in the current direction and the time difference between the time when the front of the motor vehicle in the current direction passes the stop line and the time when the rear of the motor vehicle in the current direction passes the stop line.

[0012] Further, the current direction traffic light, the current direction motor vehicle and the other direction motor vehicles in the road monitoring video of the target intersection are tracked respectively to obtain the current direction traffic light change time of the target intersection, the current direction motor vehicle movement trajectory and the other direction motor vehicle movement trajectory, and also includes: Using a convolutional neural network to perform continuity detection on the current direction traffic light, the current direction motor vehicle and the other direction motor vehicles in the road monitoring video of the target intersection; By applying a multi-target tracking algorithm, the detected current direction traffic light, current direction motor vehicle and other direction motor vehicles are tracked respectively, so as to obtain the current direction traffic light change time, current direction motor vehicle movement trajectory and other direction motor vehicle movement trajectory of the target intersection.

[0013] Furthermore, the identifying of the motion trajectories of motor vehicles in other directions of the target intersection by using the video trajectory recognition model further includes: The acquired motion trajectories of motor vehicles in other directions at the target intersection are recorded in different colors in chronological order into a continuous motion trajectory video of motor vehicles in other directions with a white background; wherein each frame of the motion trajectory video corresponds to a frame of the road monitoring video, so that the motion trajectory video is synchronized with the road monitoring video; The continuous motion trajectory videos of motor vehicles in other directions at the target intersection are identified through a video trajectory recognition model; wherein the video trajectory recognition model adopts a 3D CNN model, including several convolutional layers, pooling layers and fully connected layers; the convolutional layer is used to extract the spatiotemporal features in the continuous motion trajectory videos of motor vehicles in other directions, the pooling layer is used to reduce the size of the feature map, and the activation function adopts ReLU to increase the nonlinear ability of the 3D CNN model.

[0014] The method for detecting motor vehicles not slowing down at intersections based on road surveillance video provided by the present invention has the following advantages: it is difficult to identify the illegal behavior of motor vehicles not slowing down as required when passing through intersections, thereby providing a means for traffic management departments to supervise uncivilized traffic behaviors such as motor vehicles not giving way and not slowing down at intersections, thereby reducing the probability of traffic accidents at intersections. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. Together with the following specific embodiments, they are used to explain the present invention, but do not constitute a limitation of the present invention.

[0016] Figure 1 A flowchart of a method for detecting a motor vehicle not slowing down at an intersection based on road monitoring video provided by the present invention.

[0017] Figure 2 A flowchart of a specific implementation method of a method for detecting a motor vehicle not slowing down at an intersection based on road monitoring video provided by the present invention. DETAILED DESCRIPTION

[0018] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other. The present invention will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0019] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation method, structure, characteristics and effects of a method for detecting a motor vehicle not slowing down at an intersection based on a road monitoring video proposed by the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the protection scope of the present invention.

[0020] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so as to describe the embodiments of the present invention described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0021] In this embodiment, a method for detecting a motor vehicle that does not slow down at an intersection based on road monitoring video is provided. Figure 1 As shown, the method for detecting a motor vehicle that does not slow down at an intersection based on a road monitoring video includes: Step S1: Obtain road surveillance video of the target intersection; It should be noted that the road monitoring video of a certain direction of the intersection to be detected is accessed through the interface module for preprocessing, such as image denoising, resolution adjustment, etc.

[0022] Step S2: Track the current direction traffic light, the current direction motor vehicle and the other direction motor vehicles in the road monitoring video of the target intersection respectively, so as to obtain the current direction traffic light change time, the current direction motor vehicle movement trajectory and the other direction motor vehicle movement trajectory of the target intersection; wherein the current direction motor vehicle movement trajectory includes the current direction motor vehicle passing the stop line time; Preferably, Figure 2 As shown, the current direction traffic light, the current direction motor vehicle and the other direction motor vehicles in the road monitoring video of the target intersection are tracked respectively to obtain the current direction traffic light change time of the target intersection, the current direction motor vehicle movement trajectory and the other direction motor vehicle movement trajectory, and also includes: Using a convolutional neural network (CNN) such as ResNet and VGG, the current direction traffic light, the current direction motor vehicle and the other direction motor vehicles in the road monitoring video of the target intersection are respectively detected in continuity; By applying a multi-target tracking algorithm (such as Deep SORT), the detected current direction traffic light, current direction motor vehicle and other direction motor vehicles are tracked respectively, so as to obtain the current direction traffic light change time, current direction motor vehicle movement trajectory and other direction motor vehicle movement trajectory of the target intersection.

[0023] Specifically, (1) Tracking target initialization: After detecting motor vehicles and traffic lights in the video, the tracking target is initialized. Each target is assigned a unique ID and its initial position, velocity, acceleration and other information are recorded. (2) Kalman filter prediction: The Kalman filter is used to predict the state of each tracked target and predict the position of each target in the next frame. The Kalman filter combines the motion model of the target and the observations in the previous frame to provide a statistically based prediction result. (3) Data association: The Hungarian algorithm or GIoU (Generalized Intersection over Union) metric is used for data association to match the predicted target position with the detection results in the current frame. This process determines which detection results are associated with which tracking targets, which targets are new, and which targets may have disappeared. (4) Target trajectory update: Based on the time series of video frames, the trajectory of the tracking target is updated using a prediction algorithm such as a Kalman filter. At the same time, deep learning features (such as appearance features) are combined to correct the trajectory to avoid mistracking caused by occlusion or complex scenes. Among them, for the successfully associated targets, the detection results of the current frame are used to update the state of the Kalman filter, including information such as position and velocity. For targets that fail to match the detection results, it is assumed that they are still in the scene for a short period of time and continue to be predicted through the Kalman filter. (5) Target loss processing: When a target is lost in multiple consecutive frames, the algorithm will enter the target re-identification stage. By re-detecting the target's appearance features and matching them with the historical records, the target trajectory is attempted to be restored. Among them, if a tracked target fails to find a matching detection result in multiple consecutive frames, it is considered that the target may have left the scene and will be deleted from the tracking list. However, if the target is re-detected in subsequent frames, it can be re-associated based on its ID to resume tracking. (6) Tracking threshold and parameter setting: Set a reasonable tracking threshold (such as IoU threshold) to control the judgment conditions for target merging and splitting. Parameter settings include tracking window size, speed and acceleration update rate, etc. to ensure the robustness of tracking in complex traffic scenes. Among them, during the tracking process, the tracking loss threshold and data association matching threshold are set. For example, the matching threshold can be set based on IoU. If the IoU of two targets is lower than a certain threshold, they are not considered to be the same target. The loss threshold can be set to the number of consecutive frames in which a target is not detected before it is considered lost. (7) Result output: The final output is the motion trajectory of each tracked target in the entire video. This trajectory information will be used for subsequent analysis, such as determining whether the vehicle is not slowing down.

[0024] Step S3: According to the traffic light change time of the current direction of the target intersection, the movement trajectory of the motor vehicle in the current direction and the movement trajectory of the motor vehicles in other directions, it is determined whether the motor vehicle in the current direction does not slow down when passing the target intersection.

[0025] Preferably, Figure 2 As shown, the step S3 further includes: According to the time when the traffic light in the current direction of the target intersection changes to green or green and the time when the motor vehicle in the current direction passes the stop line, the time interval between the time when the motor vehicle in the current direction passes the stop line and the time when the red light in the current direction turns on is calculated; It should be noted that: (1) Continuous detection of traffic light status: Through continuous detection of traffic lights, the time of traffic light status changes, especially the switching time between red and yellow lights, is accurately recorded. (2) Motor vehicle trajectory tracking: Based on the initialized motor vehicle tracking target, the movement trajectory of the motor vehicle is recorded in real time, especially the time point when it passes the stop line. (3) Time matching and calculation: Combine the time of traffic light change and the time when the motor vehicle passes the stop line to calculate whether the vehicle is suspected of not slowing down. If the motor vehicle passes the stop line within 3 seconds after the yellow light comes on, it is preliminarily judged that it is suspected of not slowing down.

[0026] If the time interval between the time when the motor vehicle in the current direction passes the stop line and the time when the red light in the current direction turns on is less than the standard value, the motion trajectory of motor vehicles in other directions of the target intersection is identified through the video trajectory recognition model; among them, the standard value is generally selected to be around 3 seconds, and further analysis is required for exceeding or falling below this value. The specific standard can be adjusted according to traffic rules and the actual situation of the intersection. If the time interval is found to be too short, the vehicle is marked as a suspected vehicle that does not slow down, and the relevant data is recorded for subsequent analysis.

[0027] Specifically, the identifying the motion trajectories of motor vehicles in other directions of the target intersection by using the video trajectory recognition model also includes: The acquired motion trajectories of motor vehicles in other directions at the target intersection are recorded in different colors in chronological order into a continuous motion trajectory video of motor vehicles in other directions with a white background; wherein each frame of the motion trajectory video corresponds to a frame of the road monitoring video, so that the motion trajectory video is synchronized with the road monitoring video; It should be noted that (1) before generating the trajectory video, the original video is further processed, including video denoising, frame difference, edge enhancement, etc., to improve the visualization of the trajectory. Ensure that the trajectory in each frame of the video is clearly visible to facilitate subsequent trajectory analysis. (2) A large number of training samples are obtained from surveillance videos at different intersections, with a focus on collecting videos of vehicles that do not slow down. Use professional annotation tools to annotate these video samples, clarify the standards for vehicles that do not slow down and other abnormal behaviors, and provide a high-quality data set for model training.

[0028] The continuous motion trajectory videos of motor vehicles in other directions at the target intersection are identified through a video trajectory recognition model; wherein the video trajectory recognition model adopts a 3D CNN model, including several convolutional layers, pooling layers and fully connected layers; the convolutional layer is mainly used to extract the spatiotemporal features in the continuous motion trajectory videos of motor vehicles in other directions, and the pooling layer is mainly used to reduce the size of the feature map to avoid overfitting; the activation function adopts ReLU to increase the nonlinear ability of the 3D CNN model.

[0029] Specifically, (1) Source and annotation method of training data set: The training data set mainly comes from actual road surveillance videos and is manually annotated in detail to ensure the accuracy and representativeness of each sample. (2) Data enhancement strategy: Use data enhancement techniques such as random rotation, translation, and scaling to increase the robustness of the model. By adjusting parameters such as sample brightness and contrast, simulate the conditions under different lighting conditions. (3) Loss function selection and hyperparameter setting: Use the cross entropy loss function to optimize the classification accuracy of the model. Set a reasonable learning rate, batch size, and number of iterations to ensure the convergence and generalization ability of the model. (4) Optimization and environment of the training process: Use the Adam optimizer for model training, and accelerate the training process in a GPU acceleration environment (such as NVIDIA Tesla V100). Choose TensorFlow or PyTorch as the training framework to ensure the portability and ease of use of the model.

[0030] When it is identified that the movement trajectory of the motor vehicle in the other direction is abnormal, it is determined that the motor vehicle in the current direction is suspected of not slowing down when passing the target intersection; The speed of the motor vehicle in the current direction passing through the target intersection is calculated. When the speed of the motor vehicle in the current direction passing through the target intersection is greater than a preset value, it is determined that the motor vehicle in the current direction does not decelerate when passing through the target intersection.

[0031] Specifically, the step of calculating the speed of the motor vehicle passing through the target intersection in the current direction further includes: Using a vehicle type recognition algorithm to identify the vehicle type of the motor vehicle in the current direction in the road monitoring video of the target intersection; When the vehicle type of the motor vehicle in the current direction is identified, the actual length of the vehicle body in the current direction is obtained according to the existing vehicle type registration information; When the vehicle type of the motor vehicle in the current direction cannot be identified, the vehicle body pixel length of the motor vehicle in the current direction is extracted from the road monitoring video of the target intersection, and the actual vehicle body length of the motor vehicle in the current direction is obtained according to the vehicle body pixel length and the actual vehicle body length of the motor vehicle of the identified vehicle type in the road monitoring video of the target intersection; Specifically, (1) Implementation of vehicle model recognition algorithm: Use deep learning algorithms (such as ResNet) to train the collected vehicle model images from the rear of the vehicle. The algorithm first extracts the feature vector of the image, then classifies it through the fully connected layer, and finally outputs the vehicle model category. (2) Training process and data preparation: Collect and annotate a large number of vehicle model images from the rear of the vehicle to ensure that common vehicle models are covered. Improve the generalization ability of the model through data enhancement technology. During the training process, vehicles of the same category are used as positive samples, and vehicles of different categories are used as negative samples to ensure the model's ability to distinguish. (3) Algorithm optimization and accuracy improvement: Improve the accuracy of the model by tuning hyperparameters (such as learning rate, regularization parameters, etc.) and combining cross-validation methods. The final algorithm can effectively identify the vehicle model of suspected non-decelerating motor vehicles, and accurately obtain the vehicle body length in combination with the registration information of the vehicle management office.

[0032] It should be noted that in the same video, the motor vehicle with the identified model is selected as a reference, and the actual length of the vehicle with the unidentified model is calculated by comparing the ratio of pixel length to actual length. The camera's shooting angle and focal length are ensured to remain consistent throughout the process to ensure the accuracy of the calculation results.

[0033] The speed of the motor vehicle in the current direction passing through the target intersection is calculated based on the actual length of the motor vehicle in the current direction and the time difference between the time when the front of the motor vehicle in the current direction passes the stop line and the time when the rear of the motor vehicle in the current direction passes the stop line.

[0034] It should be noted that by comparing the speed of the motor vehicle passing through the target intersection in the current direction with the standard value of not decelerating, if the vehicle speed exceeds the set threshold, it can be further confirmed that the vehicle is suspected of not decelerating.

[0035] In an embodiment of the present invention, when a motor vehicle passes through an intersection, the trajectories of vehicles in other directions are tracked at the same time, and the motor vehicle is found not to slow down at the intersection through the abnormal trajectories of vehicles in other directions. Specifically, a multi-target tracking algorithm (such as Deep SORT) is applied to track the motor vehicles in the intersection monitoring video and the motor vehicles in other directions to obtain the vehicle trajectories. According to the abnormal situations such as sudden pauses and turns in the trajectories of vehicles in other directions, it is found that the motor vehicle is suspected of not slowing down at the intersection. Then, the speed of the motor vehicle in this direction is confirmed by measuring the speed of the video.

[0036] To ensure that the accuracy of the judgment is not affected by personal experience, the motor vehicle trajectories at the intersection are drawn in different colors on a white background to form a trajectory video (pure trajectory video without other image information). Collect and annotate a large number of video samples with motor vehicles not slowing down at intersections, and extract trajectory videos. Train a 3D CNN deep learning video trajectory recognition model for motor vehicles not slowing down at intersections. Use the video trajectory recognition model to identify intersection trajectory videos to confirm whether there are abnormal vehicle trajectories in other directions, and then determine whether there are motor vehicles not slowing down at intersections.

[0037] The video trajectory recognition model can also partially solve the problem of difficulty in accurately measuring speed based on video. After sufficient sample training, the recognition model can also identify situations where the vehicle does not slow down at the intersection based only on the trajectory of motor vehicles in this direction.

[0038] In the embodiment of the present invention, the time when the traffic light in the current direction changes and the time when the motor vehicle crosses the stop line are tracked and compared, and the suspected motor vehicle does not slow down at the intersection due to the short interval time. Specifically, "rushing the green light" is a common behavior of motor vehicles not slowing down at intersections. Not only does it not slow down, but it also accelerates. For example, if you pass the stop line about 3 seconds before the light changes (usually during the yellow light period), there is a high probability that there is no deceleration at the intersection.

[0039] In an embodiment of the present invention, a vehicle model recognition algorithm is used to obtain the actual vehicle body length to calculate the speed of a motor vehicle, specifically as follows: (1) Collect vehicle model image samples in the rear direction, train a vehicle rear model recognition algorithm based on deep learning, and obtain the vehicle body length of each vehicle model based on the existing vehicle model registration information. (2) Track and record the time when the front and rear of the motor vehicle in the same direction pass through the road monitoring video. The vehicle body length divided by the passing time can be used to obtain the approximate speed of the motor vehicle passing through the intersection. Based on this, it can also be judged whether the motor vehicle does not slow down when passing through the intersection. (3) For vehicles whose models are difficult to identify, the actual vehicle length can be obtained by comparing the pixel lengths of vehicles that can be identified in the same video. For example, if a vehicle with a vehicle length of 100 pixels is identified as a vehicle with a body length of 5m, then a vehicle with a vehicle length of 200 pixels, although the vehicle model cannot be identified, its actual body length should be 10m. With the actual vehicle length, the vehicle speed at the intersection can still be obtained by the above method, thereby judging whether the vehicle slows down when passing through the intersection.

[0040] In view of the difficulty in identifying the illegal behavior of motor vehicles failing to slow down as required when passing through intersections, the present invention provides a method for detecting motor vehicles that do not slow down at intersections based on road surveillance videos. The method adopts a multi-target video tracking algorithm based on deep learning. When a motor vehicle passes through an intersection, the trajectories of vehicles in other directions are tracked at the same time. The method detects motor vehicles that do not slow down at intersections based on trajectory characteristics supplemented by traffic light status, video-based vehicle speed estimation and other methods. This provides traffic management departments with a means of supervising uncivilized traffic behaviors such as motor vehicles not giving way and not slowing down at intersections, thereby reducing the probability of traffic accidents at intersections.

[0041] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with this profession can make some changes or modifications to equivalent embodiments of equivalent changes using the technical contents disclosed above without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the scope of the technical solution of the present invention.

Claims

1. A method for detecting a motor vehicle that does not slow down at an intersection based on road monitoring video, characterized in that: The method for detecting a motor vehicle not slowing down at an intersection based on a road monitoring video comprises the following steps: Step S1: Obtain road surveillance video of the target intersection; Step S2: Tracking the current direction traffic light, the current direction motor vehicle and the other direction motor vehicles in the road monitoring video of the target intersection respectively, so as to obtain the current direction traffic light change time, the current direction motor vehicle movement trajectory and the other direction motor vehicle movement trajectory of the target intersection; wherein the current direction motor vehicle movement trajectory includes the current direction motor vehicle passing the stop line time; Step S3: According to the traffic light change time of the current direction of the target intersection, the movement trajectory of the motor vehicle in the current direction and the movement trajectory of the motor vehicles in other directions, it is determined whether the motor vehicle in the current direction does not slow down when passing the target intersection.

2. The method for detecting a motor vehicle that does not slow down at an intersection based on road monitoring video according to claim 1, characterized in that: The step S3 further includes: According to the time when the traffic light in the current direction of the target intersection changes to green or green and the time when the motor vehicle in the current direction passes the stop line, the time interval between the time when the motor vehicle in the current direction passes the stop line and the time when the red light in the current direction turns on is calculated; If the time interval between the time when the motor vehicle in the current direction passes the stop line and the time when the red light in the current direction turns on is less than the standard value, the motion trajectory of motor vehicles in other directions of the target intersection is identified by the video trajectory recognition model; When it is identified that the movement trajectory of the motor vehicle in the other direction is abnormal, it is determined that the motor vehicle in the current direction is suspected of not slowing down when passing the target intersection; The speed of the motor vehicle in the current direction passing through the target intersection is calculated. When the speed of the motor vehicle in the current direction passing through the target intersection is greater than a preset value, it is determined that the motor vehicle in the current direction does not decelerate when passing through the target intersection.

3. The method for detecting a motor vehicle that does not slow down at an intersection based on road monitoring video according to claim 2, characterized in that: The step of calculating the speed of the motor vehicle passing through the target intersection in the current direction also includes: Using a vehicle type recognition algorithm to identify the vehicle type of the motor vehicle in the current direction in the road monitoring video of the target intersection; When the vehicle type of the motor vehicle in the current direction is identified, the actual length of the vehicle body in the current direction is obtained according to the existing vehicle type registration information; When the vehicle type of the motor vehicle in the current direction cannot be identified, the vehicle body pixel length of the motor vehicle in the current direction is extracted from the road monitoring video of the target intersection, and the actual vehicle body length of the motor vehicle in the current direction is obtained according to the vehicle body pixel length and the actual vehicle body length of the motor vehicle of the identified vehicle type in the road monitoring video of the target intersection; The speed of the motor vehicle in the current direction passing through the target intersection is calculated based on the actual length of the motor vehicle in the current direction and the time difference between the time when the front of the motor vehicle in the current direction passes the stop line and the time when the rear of the motor vehicle in the current direction passes the stop line.

4. The method for detecting a motor vehicle that does not slow down at an intersection based on road monitoring video according to claim 1, characterized in that: The method of tracking the current direction traffic light, the current direction motor vehicle and the other direction motor vehicle in the road monitoring video of the target intersection respectively, so as to obtain the current direction traffic light change time, the current direction motor vehicle movement trajectory and the other direction motor vehicle movement trajectory of the target intersection, further includes: Using a convolutional neural network to perform continuity detection on the current direction traffic light, the current direction motor vehicle and the other direction motor vehicles in the road monitoring video of the target intersection; By applying a multi-target tracking algorithm, the detected current direction traffic light, current direction motor vehicle and other direction motor vehicles are tracked respectively, so as to obtain the current direction traffic light change time, current direction motor vehicle movement trajectory and other direction motor vehicle movement trajectory of the target intersection.

5. The method for detecting a motor vehicle that does not slow down at an intersection based on road monitoring video according to claim 2, characterized in that: The step of identifying the motion trajectories of motor vehicles in other directions of the target intersection by using the video trajectory recognition model also includes: The acquired motion trajectories of motor vehicles in other directions at the target intersection are recorded in different colors in chronological order into a continuous motion trajectory video of motor vehicles in other directions with a white background; wherein each frame of the motion trajectory video corresponds to a frame of the road monitoring video, so that the motion trajectory video is synchronized with the road monitoring video; The continuous motion trajectory videos of motor vehicles in other directions at the target intersection are identified through a video trajectory recognition model; wherein the video trajectory recognition model adopts a 3D CNN model, including several convolutional layers, pooling layers and fully connected layers; the convolutional layer is used to extract the spatiotemporal features in the continuous motion trajectory videos of motor vehicles in other directions, the pooling layer is used to reduce the size of the feature map, and the activation function adopts ReLU to increase the nonlinear ability of the 3D CNN model.

Images