�ݺ�ߣ

The document discusses a gait recognition technique that uses 3D skeleton information from a Kinect sensor to track skeleton points over time. It computes covariance matrices between the skeleton point trajectories to form gait models for training and testing data. Gait recognition is performed by finding the minimum dissimilarity between these models, achieving over 90% accuracy on a dataset of 20 subjects under fixed and moving camera scenarios.