Multi-scale Deep Learning Architectures for Person Re-identification

Person Re-identification (re-id) aims to match people across non-overlapping camera views in a public space. It is a challenging problem because many people captured in surveillance videos wear similar clothes. Consequently, the differences in their appearance are often subtle and only detectable at the right location and scales. Existing re-id models, particularly the recently proposed deep learning based ones match people at a single scale. In contrast, in this paper, a novel multi-scale deep learning model is proposed. Our model is able to learn deep discriminative feature representations at different scales and automatically determine the most suitable scales for matching. The importance of different spatial locations for extracting discriminative features is also learned explicitly. Experiments are carried out to demonstrate that the proposed model outperforms the state-of-the art on a number of benchmarks.

1. Introduction

Person re-identification (re-id) is defined as the task of matching two pedestrian images crossing non-overlapping camera views [11]. It plays an important role in a number of applications in video surveillance, including multi-camera tracking [2], [41], crowd counting [3], [10], and multi-camera activity analysis [54], [53]. Person re-id is extremely challenging and remains unsolved for a number of reasons. First, in different camera views, one person's appearance often changes dramatically caused by the variances in body pose, camera viewpoints, occlusion and illumination conditions. Second, in a public space, many people often wear very similar clothes (e.g., dark coats in winter). The differences that can be used to tell them apart are often subtle, which could be the global, e.g., one person is bulkier than the other, or local, e.g., the two people wear different shoes.