Validity of the microsoft kinect system in assessment of compensatory stepping behavior during standing and treadmill walking

Abstract

Background: Rapid compensatory stepping plays an important role in preventing falls when balance is lost; however, these responses cannot be accurately quantified in the clinic. The Microsoft Kinect™ system provides real-time anatomical landmark position data in three dimensions (3D), which may bridge this gap. Methods: Compensatory stepping reactions were evoked in 8 young adults by a sudden platform horizontal motion on which the subject stood or walked on a treadmill. The movements were recorded with both a 3D-APAS motion capture and Microsoft Kinect™ systems. The outcome measures consisted of compensatory step times (milliseconds) and length (centimeters). The average values of two standing and walking trials for Microsoft Kinect™ and the 3D-APAS systems were compared using t-test, Pearson’s correlation, Altman-bland plots, and the average difference of root mean square error (RMSE) of joint position. Results: The Microsoft Kinect™ had high correlations for the compensatory step times (r = 0.75–0.78, p = 0.04) during standing and moderate correlations for walking (r = 0.53–0.63, p = 0.05). The step length, however had a very high correlations for both standing and walking (r > 0.97, p = 0.01). The RMSE showed acceptable differences during the perturbation trials with smallest relative error in anterior-posterior direction (2-3%) and the highest in the vertical direction (11–13%). No systematic bias were evident in the Bland and Altman graphs. Conclusions: The Microsoft Kinect™ system provides comparable data to a video-based 3D motion analysis system when assessing step length and less accurate but still clinically acceptable for step times during balance recovery when balance is lost and fall is initiated.