HUMAN POSE RECOGNITION USING GEODESIC DISTANCE AND COLOR FEATURES WITH DEPTH CAMERA

Abstract

The paper presents an approach to recover a full-body 3D human pose using geodesic and color features captured by a depth camera. The 3D information obtained from the depth images is employed to represent the points belonging to a human body in the form of a graph. The interest points or landmark locations with definite geodesic distances from the human body centroid are extracted to locate areas of the hand, foot, and face based on color images using a skin detection algorithm. Utilizing the anatomical landmark locations, joint angles of the body parts are computed. The estimated joint angles are then mapped to the body parts of a 3D human body model, which consists of a set of connected parts. Finally, the 3D human model reflects the human pose estimate. In our experiments, we assessed the detection of anatomical landmarks by our algorithm and then the 3D pose recovery for simple hand gestures, complex hand gestures, simple full body movements and complex body movements. The qualitative results of pose approximations depict the fact that the presented methodology is efficient enough to get good estimates of various full body movements.