Human Identification Recognition in Surveillance Videos
Kai Jin, Xuemei Xie, Fangyu Wang, Xiao Han, Guangming Shi · 2019
Although face recognition has achieved good performance in attendance marking system, people need to cooperate with the camera. Human identification recognition performs poorly in surveillance videos because of the deformed and occluded conditions. In this paper, we propose a new method to recognize human identification using global and local structural information. Firstly, we combine pedestrian detection and tracking with face recognition, in order to improve the recognition performance of occlusion and deformed faces. Secondly, we propose a selective recognition algorithm based on pedestrian trajectory, which is used to identify the pedestrian walking towards the webcam. Finally, we conduct experiments on our own human identification dataset, which contains 93 challenging video sequences captured in the corridor. Our method achieves better performance than methods only using face information.