3D Keypoint Tracking Based on Hybrid Flow Computation for Human Motion Analysis

In this paper, we propose a robust method to track 3D keypoints for representing 3D human motions under noisy depth environment. Once keypoints are extracted for tracking, the subsequent locations of the keypoints are found by estimating their 3D motion flows. However, estimating the flows using depth information would be a problem, especially, when some depth values are unavailable or the data points are severely affected by noise. To handle such circumstances, we suggest hybrid flow computation scheme that selectively utilizes range flow and optical flow estimation methods. Experimental results show that the proposed method outperforms the conventional flow estimation methods under problematic environment in terms of tracking accuracy and performance, with linear increase in computational time.