Timmi Alessandro, Coates Gino, Fortin Karine, Ackland David, Bryant Adam L, Gordon Ian, Pivonka Peter
St Vincent's Department of Surgery, The University of Melbourne, 29 Regent St, Fitzroy, VIC 3065, Australia.
Centre for Health, Exercise & Sports Medicine, The University of Melbourne, 202-206 Berkeley St, Carlton, VIC 3053, Australia.
Med Eng Phys. 2018 Sep;59:63-69. doi: 10.1016/j.medengphy.2018.04.020. Epub 2018 Jul 6.
Microsoft Kinect for Windows v2 is a motion analysis system that features a markerless human pose estimation algorithm. Given its affordability and portability, Kinect v2 has potential for use in biomechanical research and within clinical settings; however, recent studies suggest high inaccuracy of the markerless algorithm compared to marker-based motion capture systems. A novel tracking method was developed using Kinect v2, employing custom-made colored markers and computer vision techniques. The aim of this study was to test the accuracy of this approach relative to a conventional Vicon motion analysis system, performing a Bland-Altman analysis of agreement. Twenty participants were recruited, and markers placed on bony prominences near hip, knee and ankle. Three-dimensional coordinates of the markers were recorded during treadmill walking and running. The limits of agreement (LOA) of marker coordinates were narrower than - 10 and 10 mm in most conditions, however a negative relationship between accuracy and treadmill speed was observed along Kinect depth direction. LOA of the surrogate knee angles were within - 1.8°, 1.7° for flexion in all conditions and - 2.9°, 1.7° for adduction during fast walking. The proposed methodology exhibited good agreement with a marker-based system over a range of gait speeds and, for this reason, may be useful as low-cost motion analysis tool for selected biomechanical applications.
微软Kinect for Windows v2是一种运动分析系统，具有无标记人体姿态估计算法。鉴于其价格实惠且便于携带，Kinect v2在生物力学研究和临床环境中有应用潜力；然而，最近的研究表明，与基于标记的运动捕捉系统相比，无标记算法的误差较大。利用Kinect v2开发了一种新颖的跟踪方法，采用定制的彩色标记和计算机视觉技术。本研究的目的是相对于传统的Vicon运动分析系统测试这种方法的准确性，进行一致性的布兰德-奥特曼分析。招募了20名参与者，并在髋、膝和踝关节附近的骨突处放置标记。在跑步机行走和跑步过程中记录标记的三维坐标。在大多数情况下，标记坐标的一致性界限（LOA）窄于-10和10毫米，然而，沿Kinect深度方向观察到准确性与跑步机速度之间呈负相关。在所有条件下，替代膝关节角度的LOA在屈曲时为-1.8°至1.7°，在快走时内收时为-2.9°至1.7°。所提出的方法在一系列步态速度下与基于标记的系统表现出良好的一致性，因此，对于选定的生物力学应用，可能作为低成本的运动分析工具。
