Pedestrian re-ID based on deep feature

Due to variations in pose and illumination, the backgrounds are various in different cameras. The change of the background plays a decisive role in pedestrian re-id. This paper provides a method of pedestrian re-id to adopt two deep features based on whether backgrounds interfere with pedestrians. First of all, we divide the sample of pedestrian into three parts, the head, the body and the legs. We determine whether there is background interference by comparing the HSV distance between the background and the body image. Finally, different deep features are selected by comparing with thresholds. Our experimental results on two datasets show that the proposed method outperforms the state-of-art approaches.