Human Joint Skeleton Tracking Using Multiple Kinect Azure

Development of human skeleton tracking requires high accuracy with affordable installation to enlarge their applicability. This paper proposes human skeleton tracking using multiple cameras to minimize occlusion. We combined multiple human tracking modules from Azure based on ONNX Runtime. Each module uses passive infrared as the human segmentation and recognition process and depth information to transform to the 3D position. Each module’s output skeleton data is separated into six parts: right-upper limb, left-upper limb, right-lower limb, left-lower limb, head, and torso. Then those part is compared and combined based on the confidence evaluation based on the occlusion probabilistic. Furthermore, 35 joint skeleton angles are also generated based on the anatomy of the human musculoskeletal model. The effectiveness of the proposed model has been validated from the comparison with related work and some applications of human muscle activity monitoring.