People’s Fast Moving Detection Method in Buses Based on YOLOv5
Zhang Xiao-ping, Ji Jiahui, Wang Li, Zhonghe He, Liu Shida
Abstract
To ensure the public’s safety such as in buses, it is very important to accurately judge people’s behaviors and give early warnings. If by watching the video surveillance manually, the cost will be very high, and it cannot be effectively popularized, so video automatic monitoring is preferred. For buses, its environmental space is closed as well as narrow, and at the same time, it is often in a non-stationary state, so traditional behavior detection methods cannot be used here as they are easily affected by moving environment and difficult to fulfill object behavior identification in real time. Aiming at this problem, for people’s fast-moving in buses, a kind of detection method based on YOLOv5 is proposed in this paper. Firstly, the method detects people through one-stage object detection. Secondly, in order to obtain the person's movement trajectory quickly and accurately, an improved two-stage object matching algorithm is designed to track different people. Then, the speed curves of a person during normal activities and fast moving are compared. Finally, an abnormal alarm mechanism is constructed to realize the effective fast movement alarm. Surveillance video in the bus was used to test and evaluate the effectiveness of the method. Results show that the accuracy rate of our method can get 95.4%.