Research And Application Of Vehicle Trajectory Anomaly Detection Algorithm In The Highway Scene

Real-time detection of abnormal vehicle trajectories is crucial to ensure the safety of highway traffic.Currently,highway management companies have established a relatively complete video surveillance system.Through real-time analysis and processing of video surveillance data and using artificial intelligence technology,the detection of abnormal vehicle trajectories has significant application value.However,in the highway scene,there are problems such as small target vehicle omission,multi-scale changes,tracking trajectory interruption,and low real-time performance,which result in high missed detection rate and low real-time performance of vehicle trajectory abnormal detection algorithms.To address these issues,this paper studies the YOLOv5 s detection algorithm and Bo T-SORT tracking algorithm and sets a discrimination method to achieve the detection of abnormal vehicle trajectories,reducing the false positive rate and missed detection rate.The main research contents of this paper are as follows:(1)An improved lightweight vehicle detection algorithm MG-YOLOv5 s based on YOLOv5 s was proposed.In the highway scene,there are problems such as large scale changes between distant and near vehicles,easy missed detection of small target vehicles in the distance,and poor robustness of the detection model.The vehicle detection model has a large number of parameters and slow detection speed,requiring faster detection speed to meet the real-time requirements of vehicle tracking in subsequent videos.To address these issues,an improved Mobile Netv3 lightweight network was first used as the backbone network to reduce the model’s parameter count and improve detection speed.Then,a weighted bidirectional feature pyramid fusion structure was used to utilize feature information between different scales,combined with Ghost modules to reduce the impact of scale and enhance the detection ability of small target vehicles.Finally,the ECA module was introduced to improve detection accuracy.Experimental results show that the m AP of the MG-YOLOv5 s algorithm reaches 90.7%,with an average detection time of 17.4ms,effectively improving the speed of vehicle detection and the detection rate of small target vehicles.(2)An improved vehicle tracking algorithm A-Bo T-SORT based on Bo T-SORT was proposed.When tracking vehicles traveling on highways,there are problems such as low real-time performance during tracking,mismatch between detection boxes and predicted tracking boxes leading to identity ID jumps,and discontinuous vehicle trajectories.To address these issues,the AKAZE feature detection description algorithm was first used to match the detection box with adjacent images through local matching features,combined with the Kalman filter predicted box to obtain the vehicle box position,reducing identity jumps and improving tracking effectiveness.Then,the MG-YOLOv5 s algorithm was used as the detector to improve the detection effect and reduce the overall tracking time.Experimental results show that the A-Bo T-SORT algorithm combined with the MG-YOLOv5 s algorithm has varying degrees of improvement in evaluation indicators,reduces identity ID jumps,reaches a frame rate of 29 frames per second(FPS),and improves overall tracking speed and performance.(3)An algorithm for detecting abnormal vehicle trajectories was designed and implemented,and applied to a highway vehicle abnormal event detection system.The vehicle trajectory anomaly detection algorithm is based on the vehicle tracking to obtain the position of the vehicle trajectory.The algorithm uses mean filtering to smooth the vehicle trajectory sequence and then sets a discrimination method based on the relationship between time and trajectory position to achieve vehicle trajectory anomaly detection.The algorithm model can effectively detect two typical types of vehicle anomalies: abnormal parking and vehicle retrograde trajectory,reducing false alarms and missed detections.When applying this algorithm model to the highway vehicle anomaly event detection system,the results show that the system can meet the real-time and accuracy requirements for anomaly event detection in the highway scene.Figure [29] Table [11] Reference [71]...