Multiscale feature fusion for surveillance video diagnosis

Fanglin Chen, Weihang Wang, Huiyuan Yang, Wenjie Pei, Guangming Lu

Abstract: Recently, surveillance video diagnosis has attracted increasing interest for generating real-time alarms related to camera failure in video surveillance systems. The existing surveillance video diagnosis methods do not have sufficient ability to detect multiple types of anomalies. Therefore, this paper proposes a surveillance video diagnosis method based on deep learning to detect multiple types of anomalies. A multiscale feature fusion residual network is designed to detect and classify camera anomalies. The experimental results show that the classification accuracy of the proposed method is more than 98%.