Pedestrian Re-Identification Based on Gait Analysis

Pedestrian re-identification is a crucial task in various safety applications, such as traffic management, collision avoidance, and emergency response. One of the challenging issues in pedestrian re-identification is how to identify the same object when the collection perspective changes or carrying different items. To address the above issues, this paper proposes a new gait based pedestrian re-identification method from the perspective of improving the representation ability of different pedestrians. The proposed method extracts spatial and temporal features separately from the gait data and designs a spatiotemporal feature fusion module at the top of the network to prevent features from interfering with each other in different domains. The temporal scoring strategy based on gated recurrent unit (GRU) is then used to evaluate the image quality of each frame in the network branch where temporal gait features are extracted. Finally, the proposed method is evaluated on two open datasets, CASIA-B and OU-ISIR-LP. Experimental results show that, compared with existing methods, the proposed method has a higher recognition rate under cross-view and scene conditions, such as carrying a bag and heavily clothed, in which the proposed method demonstrates better performance in pedestrian recognition. These results highlight the effectiveness and robustness of the proposed method and demonstrate its potential as a future research direction in pedestrian re-identification.