Abnormality Monitoring and Recognition of Surveillance Video Based on ResNet Residual Network
Hao Yang, Deng Chen, Xiancheng Feng · 2024
In order to solve the problems of high cost of manual identification of abnormal situations in video surveillance and poor generalization ability of traditional algorithms, this paper proposes an anomaly detection and recognition algorithm for surveillance video based on ResNet residual network. First, the feature representation of the surveillance video frame is extracted through ResNet-18 to support the subsequent anomaly detection task. Secondly, he is introduced_ Normal method is used to initialize the network weight of ResNet-18 to speed up the network training process. Finally, the SGD optimization algorithm is combined with ResNet-18 to improve the accuracy and efficiency of anomaly detection. The experimental results show that the method has achieved good detection and recognition results on multiple data sets, indicating that ResNet-18 has a good performance in video anomaly detection tasks. This method can be widely used in security, industrial manufacturing, transportation and other fields, improve the intelligent level and management efficiency of video surveillance system, and make contributions to social security and development.