Person Re-identification Based on Adaptive Feature Selection

Abstract

In video surveillance, it is very important to identify individuals or determine whether a given individual has already appeared over a large-scale network of cameras, which is so called problem of person re-identification. In general, the human appearance obtained in one camera is usually different from others obtained in other cameras, because of variations in view angle, illumination, body poses, clothing, background clutter and occlusion. Thus, almost none of the existing methods of person re-identification can work with satisfied accuracy. In order to address this problem, we propose a person re-identification approach by using adaptive feature selection method. Specifically, at first, we detect human and human body parts. Then, we extract certain features on each part adaptively driven by their unique and inherent appearance attributes. At last, we map the person re-identification problem to a distance learning problem, and find out the similarity between corresponding body parts. The approach has been evaluated through extensive experiments, and the results show that our method can improve the accuracy of person re-identification greatly.