On the accuracy of low-cost motion capture systems for range of motion measurements

In this paper, we study low-cost motion tracking systems for range of motion (RoM) measurements in the tele-rehabilitation context using Augmented Reality. We propose simple yet effective extensions of the Microsoft Kinect SDK 2.0 skeleton tracking algorithm. Our extensions consist of temporal smoothing of the joint estimates as well as an intuitive, patient-specific adjustment of the bone lengths that is implemented as a quick, one-Time calibration performed by the therapist. We compare our system to the Kinect v1, the non-modified Kinect v2, a marker-based optical tracking system, and the clinical gold standard set by two subject-matter-experts using a goniometer. We study the accuracy of all systems in RoM measurement on the elbow joints. We quantitatively compare angular deviation from the expert measurements and perform analysis on statistical confidence. The results indicate, that the proposed personalized setup substantially outperforms all competing systems and effectively corrects for the systematic error of the skeleton tracking, particularly at full flexion. The improved system matched the observations of both experts with a mean error of 3:78° We conclude, that the proposed, personalized method for RoM measurement with Augmented Reality feedback is promising for tele-rehabilitation scenarios. Future work will investigate whether similar strategies can be applied to more complex joints, such as the shoulder.