RGB-D Camera based Human Limb Movement Recognition and Tracking in Supine Positions

Human action recognition is an important topic and it has been widely applied in human-computer interaction. This work aims at human pose recognition based on RGB-D camera data streams for patients in a supine position (for example, with limb movement disorders), where bed backgrounds are misinterpreted as part of the human body. The target human limb poses include both upper and lower limbs’ movements. A novel framework for human limb movement recognition in a supine position is investigated and developed. The main steps include human region locating, limb detection and segmentation, limb labeling and limb movement tracking. Based on the tracking information on human bones by the RGB-D camera, human body region detection is executed according to the depth information. Further, this system is able to solve the problem of the human limb label confusions for left and right limbs due to their occlusion cases. Our proposed tracking algorithm achieves accurate positioning of the human body, and it provides reliable mid-level data for tracking human limbs. Experiment results show that four human limbs in supine positions are able to be located successfully. It is valuable and promising in the field of medical rehabilitation.