Anomaly object detection in urban management surveillance video based on deep learning

In modem city management, the detection and identification of abnormal objects is particularly important. Random parking of non-motor vehicles is one of the common important problems in the process of smart city management. The traditional processing method is mainly to find abnormal targets manually and equipment, and then report to the urban management center for the next step. These methods are time consuming, laborious, and inefficient. In order to further improve the efficiency of urban management, this paper proposes an improved YOLOv3 urban management surveillance video anomaly object detection algorithm. On the basis of the original YOLOv3 framework, we use depthwise separable convolution to reduce the amount of model calculations and increase the detection speed. At the same time, the SE module is added to improve the network structure complexity and detection accuracy. Based on the actual surveillance video data, the detection speed of this algorithm is 4 fps faster than the traditional YOLOv3, and the recognition accuracy is improved by 3%.