Asynchronous Gaussian Filtering Fusion for Human Motion Estimation Based on RGB-D Cameras

This article studies the asynchronous Gaussian filtering fusion (AGFF) problem of 3-D human motion estimation (HME) for human-machine cooperation based on RGB-D cameras, and a flexible filtering structure is proposed for integrating asynchronous measurements with a prior motion model and compensating for measurement uncertainties implicitly. Considering the complexity for capturing human pose from the color images, the measurements from the depth camera are introduced to correct prior estimates periodically, while the measurements from the color camera are generated by an event-triggered strategy. To improve the robustness against misrecognition, the measurements are fused progressively and sequentially in the AGFF method. Moreover, under the filtering structure, a new detection and compensation mechanism for misrecognition is designed to improve the filtering performance by using color-based measurements. Finally, the effectiveness of the proposed AGFF method is verified through simulations and experiments.