Comparative study between color texture and shape descriptors for multi-camera pedestrians identification

Abstract

In this paper, we propose a comparative study between different descriptors based on color, texture and shape information. In particular, our study is focused on measuring the robustness of these descriptors in terms of identifing a person in a camera network. We prove through experimental study based on VIPeR pedestrians images dataset and “Cumulative Matching Characteristic” (CMC) measurement that color descriptors are the most appropriate in multi-camera context: they are less sensitive to the highly articulated human body, changes in lighting conditions and large pose variations.