Temporal Related Attention for Video-Based Pedestrian Attribute Recognition

Pedestrian attribute recognition is an important task in the area of video surveillance. In this paper, we propose a temporal related attention(TRA) method for video-based pedestrian attribute recognition. The proposed method can automatically learn both spatial and temporal features from input image sequences. Furthermore, the proposed method can capture affinities between embedding and semantic guidance learned from Recurrent Neural Network(RNN)so that attention mask could be generated by these correlation vectors with several convolution layers. Experiments conducted on MARS and Duke-MTMC datasets show that the proposed method performs favorably against the existing approaches.