A Novel Model for Outdoor Abnormal Gait Recognition Based on Multi-View Kinect Data and Graph Convolution

To effectively improve the accuracy of outdoor abnormal gait recognition, this study proposes a novel model for abnormal gait recognition based on multi-view Kinect data and graph convolutional networks (GCN). By integrating view-invariant learning in manifold space with graph convolutional topology learning, the model can efficiently consolidate multi-view skeletal abnormal gait data. Leveraging a hierarchical spatiempotoral graph convolutional fusion network, it can effectively mine the most distinctive fine-grained abnormal gait features embedded in multi-view skeletal data, thereby significantly improving the generalization performance of the abnormal gait classification model. In this study, multi-view Kinect data simulating various abnormal gaits were collected from 30 participants to evaluate the effectiveness of the proposed model. The results indicate that the model can fully integrate multi-view Kinect abnormal gait data and effectively capture the fine-grained features that signify significant differences in abnormal gait patterns. With a classification accuracy of 92.75%, the model demonstrates markedly superior generalization performance compared to models from recent related studies, offering a new technological solution for identifying outdoor abnormal gaits.