An Approach towards Highway Abnormal Parking Detection Based on Video Spatio-Temporal Relationship
Ying Guo, Ruilin Liang, Runmin Wang · 2023
In the highway environment, it is of great significance to detect abnormal vehicles accurately and efficiently to improve highway traffic safety. Aiming at the problems of low efficiency, high miss ratio of detection, and poor real-time performance of traditional highway abnormal detection methods, a highway abnormal parking detection method based on spatio-temporal relationship is proposed. Firstly, the traffic flow frequency is used to quantify the segmented image and remove the smaller connected domain to extract the road surface information. Then the perspective geometry of the pinhole camera is normalized to the same scale through the perspective relationship model, and the YOLOv4 network is input for secondary detection to enhance the robustness of near and far target detection. Finally, the spatio-temporal information matrix is established, the merged anomaly region is detected, and the detection results are calculated by updating the spatio-temporal matrix. The experimental results show that the accuracy of this method is 95% in remote scene and 93% in crowded scene.