Person Re-Identification by Pose Invariant Deep Metric Learning With Improved Triplet Loss

• Main Authors: Min Chen, Yongxin Ge, Xin Feng, Chuanyun Xu, Dan Yang
• Format: Article
• Language: English
• Published: IEEE 2018-01-01
• Series: IEEE Access
• Subjects: Person re-identification; metric learning; multiview learning; triplet loss function
• Online Access: https://ieeexplore.ieee.org/document/8522036/

Person re-identification (re-ID) is a challenging problem in the community which aims at identifying person in a surveillance video. Despite recent advance in the field of computer vision, person re-ID still presents great challenge since person’s presence is various under different illumination, viewpoints, occlusion, and background clutter. In this paper, to exploit more discriminative information of person’s appearance, we propose a novel pose invariant deep metric learning (PIDML) method under an improved triplet loss for person re-ID. Our approach contributes the misalignment problem and distance metric simultaneously, which are two key problems for person re-ID. Extensive experiments show that our proposed method could achieve favorable accuracy while compared with the state-of-the-art techniques on the challenging Market-1501, CUHK03, and VIPeR datasets.