End-to-End Model-Based Gait Recognition with Matching Module Based on Graph Neural Networks

Existing gait recognition systems have achieved success by extracting robust gait features from silhouette images, but gait can be sensitive to appearance features such as clothing and carried items. As a more promising alternative, model-based gait recognition is robust against some variations, such as clothing and baggage carried. With the recent development of human pose estimation, the difficulty of implementing model-based methods has been mitigated. This paper introduces an end-to-end model-based gait recognition method designed for large-scale and uncontrolled datasets. The proposed method takes 3D data as input and utilizes the ST-GCN as the embedding module. We have replaced the simple nearest neighbor algorithm for better performance. Specifically, the 3D skeletons are embedded into the Graph Neural Network to represent similarities. The proposed method is evaluated on the GREW dataset, showing state-of-the-art (SOTA) results in model-based gait recognition.