Learning Hierarchical and Efficient Person Re-Identification for Robotic Navigation
Zhang Jiangning,Xu Chao,Zhao Xiangrui,Liu Liang,Liu Yong,Yao Jinqiang,Zhejiang Chipkong Technology
DOI: https://doi.org/10.1007/s41315-021-00167-2
IF: 1.7
2021-01-01
International Journal of Intelligent Robotics and Applications
Abstract:Recent works in the person re-identification task mainly focus on the model accuracy while ignoring factors related to efficiency, e.g., model size and latency, which are critical for practical application. In this paper, we propose a novel H ierarchical and E fficient Net work (HENet) that learns hierarchical global, partial, and recovery features ensemble under the supervision of multiple loss combinations. To further improve the robustness against the irregular occlusion, we propose a new dataset augmentation approach, dubbed random polygon erasing, to random erase the input image’s irregular area imitating the body part missing. We also propose an E fficiency S core (ES) metric to evaluate the model efficiency. Extensive experiments on Market1501, DukeMTMC-ReID, and CUHK03 datasets show the efficiency and superiority of our approach compared with epoch-making methods. We further deploy HENet on a robotic car, and the experimental result demonstrates the effectiveness of our method for robotic navigation.