This paper proposes a human gait recognition method based on a novel adaptive dynamics learning approach (i.e., deterministic learning) and the Kinect depth camera. First, a series of real-time human joint position data is captured and collected by using Microsoft Kinect and Robot Operating System (ROS) packages. Second, the collected position data is converted to four time-varying features that could best represent the human gait dynamics. Third, the human gait dynamics associated with the four features from each gait pattern are effectively modeled using a novel discrete-time adaptive dynamics learning algorithm. A gait dynamics database is developed, based on which a multi-model recognition scheme is implemented to realize real-time rapid recognition of human gait. Finally, a Python-based GUI is created by integrating all the algorithmic components to facilitate user operations and visualizations. In the experiment study, by using the 2-fold, 10-fold cross-validation, and leave-one-out validation approaches, the recognition success rates are reported to be average 93.7% based on our proposed method, which outperforms other state-of-the-art machine learning approaches.

This content is only available via PDF.
You do not currently have access to this content.