Detection of Abnormal Behavior Based on the Scene of Anti-photographing

Wei Zhang, Fan Lin

Abstract: In the era of increasingly prominent information security issues, Anti-Photographing is a subject worthy of study, which is based on the problem of camera equipment detection and camera behavior analysis. Although the current target detection algorithm is relatively mature, there are still problems in the accuracy of target detection in complex scenes. Resnet50 is a neural network framework with high learning efficiency based on the residual module, which is widely used in object recognition tasks. In this paper, we propose a new detection framework, which is based on the Gaussian skin color model combined with the Openpose method to improve the accuracy of hand region extraction. At the same time, the mobile phone recognition model based on the Resnet50 network is used to identify whether the hand holds a mobile phone. At last, the gestures of the person’s hand and face are combined to complete the evaluation of whether there is a photographing behavior. The results of testing experiments on the photographing behavior in the video frame show that the framework has a good detection effect on photographing behavior.

External IDs:dblp:conf/icic/ZhangL20