Using a single Kinect device for human skeleton tracking and motion tracking lacks of reliability required in sports medicine and rehabilitation domains. Human joints reconstructed from non-standard poses such as squatting, sitting and lying are asymmetric and have unnatural lengths while their recognition error exceeds the error of recognizing standard poses. In order to achieve higher accuracy and usability for practical smart health applications we propose a practical solution for human skeleton tracking and analysis that performs the fusion of skeletal data from three Kinect devices to provide a complete 3D spatial coverage of a subject. The paper describes a novel data fusion algorithm using algebraic operations in vector space, the deployment of the system using three Kinect units, provides analysis of dynamic characteristics (position of joints, speed of movement, functional working envelope, body asymmetry and the rate of fatigue) of human motion during physical exercising, and evaluates intra-session reliability of the system using test–retest reliability metrics (intra-class correlation, coefficient of variation and coefficient of determination). Comparison of multi-Kinect system vs single-Kinect system shows an improvement in accuracy of 15.7%, while intra-session reliability is rated as excellent.

使用单个 Kinect 设备进行人体骨骼跟踪和运动跟踪缺乏运动医学和康复领域所需的可靠性。由蹲、坐、卧等非标准姿势重建的人体关节不对称、长度不自然，其识别误差超过了标准姿势的识别误差。为了在实际的智能健康应用中实现更高的准确性和可用性，我们提出了一种用于人体骨骼跟踪和分析的实用解决方案，该解决方案对来自三个 Kinect 设备的骨骼数据进行融合，以提供对象的完整 3D 空间覆盖。本文描述了一种使用向量空间中的代数运算的新颖数据融合算法，使用三个 Kinect 单元部署系统，提供动态特性分析（关节位置、运动速度、功能工作范围、身体不对称性和疲劳率） ）在体育锻炼期间人体运动，并使用重测可靠性指标（类内相关性、变异系数和确定系数）评估系统的会话内可靠性。多 Kinect 系统与单 Kinect 系统的比较表明，准确度提高了 15.7%，而会话内可靠性被评为优秀。