Objective: Evaluation of Microsoft Kinect V2 against standard motion capture.

Background: Augmentation of clinical Motor testing with instrumental assessment has been proposed to increase objectivity. Previous work on the first version of the low-cost Kinect sensor for clinical use reported mixed results. The Kinect V2 improves on the first version with respect to depth resolution and motion tracking capability.

Methods: We used Motognosis Labs in combination with Microsoft KinectV2 and Software Development Kit (SDK) version 2.0. As reference motion tracking system, we used a 16-camera Vicon system (MX13+, Nexus 2.1) with the standard gait model. 19 healthy volunteers six performed six different motor tasks with 3 to 5 repetitions each. We mapped the 3D Vicon marker positions to those of the Kinect joints and resampled them to 30Hz. All movements were temporally and spatially aligned. Spatial accuracy of the Kinect joint movements is reported as Pearson correlation coefficients (p) against Vicon markers and signal to noise ratios (SNR) in each spatial dimension.

Results: Kinect accuracy was moderate to excellent. Averaged over all joint movements in the 3 spatial dimensions it was highest in foreaft (z) direction (N =11163; p_x = 0.80; p_y = 0.52; p_z = 0.93). Accuracy varied with joint location, with head joint having the highest and foot joints the lowest correlations with Vicon data (p_footL/R_z 0.64/6; p_head_z 0.99). Furthermore, spatial accuracy was found to depend on the measured task, e.g. the SpineBase joint in quiet stance had p_ x = 0.95, but p_ x = 0.64 in a postural transition task. The signal to noise ratios provided information about Kinect V2 joint stability during assessments and may serve as criterion for data cleaning.

Conclusions: We present data on the spatial accuracy of the Kinect V2 sensor in movement tracking of body joints at different locations. These data are a prerequisite to further develop this tool into clinical applications. Accuracy was found to depend on the direction of joint movement, location of skeletal joint and type of motor task. Best results were achieved for movements in AP direction of uncovered axial joints (SpineBase, SpineMid, SpineShoulder and Head). Furthermore, our data help to standardize quality control of Kinect V2 data before parameter extraction.

« Back to 2016 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/spatial-accuracy-and-reliability-of-the-microsoft-kinect-v2-in-the-assessment-of-joint-movement-in-comparison-to-marker-based-motion-capture-vicon/