Subject Centric Group Feature for Person Re-Identification

Abstract

This paper presents a subject centric group feature for person re-identification. Our approach is inspired by the observation that people often tend to walk alongside others or in a group. We argue that co-travelers' information, including geometry and visual cues, can reduce the re-identification ambiguity and lead to better accuracy, compared to approaches that rely only on visual cues. We introduce person-group feature to capture both geometry and visual information of co-travelers around a subject. We compute the dis-similarity between person-group features by solving an integer programming problem. The proposed approach is evaluated in its ability to improve the accuracy of re-identification of people traveling within groups. The results show that our approach outperforms state-of-the-art visual based as well as group information based methods.