A Spatial-Temporal Information based Traffic-Flow Detection Method for video Surveillance

Traditional video-based traffic flow detection algorithms face challenges in extracting multiple vehicle features. They also cannot properly handle situations, such as vehicle occlusion. In this case, an improved vehicle detection algorithm was proposed that integrates time and space features, leading to better accuracy. Specifically, the proposed algorithm uses the normalized values of the cross-correlation between the foreground and background image blocks as the features of the detected vehicles passing by. The spatio-temporal accumulation is applied for processing and analyzing the extracted feature values. Moreover, the deep learning target detection algorithm is introduced to update the background, thereby eliminating missed detection caused by vehicle occlusion. The results show that the proposed method is less affected by the environment than the traditional background subtraction method. Therefore, better accuracy is achieved.