Distance aggregation for person re-identification using simulated annealing algorithm

Abstract

The aim of person re-identification is to match pedestrians which across disjoint camera views. Many features have been proposed to improve the re-identification accuracy. However, due to significant person appearance variations in viewpoints, poses, and illumination across different cameras, individual feature is less discriminative to represent the different person images. In this paper, we propose an effective and easy-to-apply distance aggregation method to combine different features. The individual distance are firstly obtained by metric learning. Then we use simulated annealing algorithm to learn different distance weight. Experimental results demonstrate that the proposed method significantly outperforms the existing methods in VIPeR dataset.