Performance analysis of a generalized motion capture system using microsoft kinect 2.0

Abstract This work presents a fine-grained analysis of the performance and limitations of the Microsoft Kinect sensor for tracking human movement in the context of biomechanical research and clinical applications. Earlier work in this field has focused on scalar summary measures or ad-hoc metrics with respect to specific movements that do not generalize well across clinical applications. In this work, the performance of the Microsoft Kinect is compared to motion tracking from a concurrently sampled professional grade Qualisys motion capture system. Subjects performed a range of clinically relevant tasks such as Sit-to-Stand and Timed Up-and-Go. Captured data included both three-dimensional joint center displacements and joint angles as recorded from both systems. Kinect performance was measured using cross correlation coefficients (CCR), root mean squared error (RMSE) relative to the Qualisys gold-standard and a new summary metric (SM) that combines both. Our results show that the Kinect-based system provides adequate performance when tracking joint center displacements in time, with overall CCR = 0.78, RMSE = 3.35 cm and SM = 1.21. On the contrary, lower accuracy was measured when tracking joint angles, with CCR = 0.58, RMSE = 24.59°, and SM = 3.76. Although performance differences for various movements and motion planes have been found, the results suggest that the Kinect is a viable tool for general biomechanical research, with specific limits on what levels of performance can be expected under various conditions.