Human skeleton tracking using information weighted consensus filter in distributed camera networks

Abstract

Human body estimation is one of the active research fields in computer vision. It is widely used in human-computer interaction, pattern recognition, intelligent monitoring system and human behavior detection. A single depth camera has a limited range of observations thus can only estimate the human joint position which is in its field of view. To track human motion in a large area, a camera network can be employed. The information weighted consensus filter (ICF) is a distributed estimation algorithm which has a good estimation effect in a camera network with limited sensing range. The method proposed in this paper introduces the information weighted consensus filter to track the joint points of skeleton separately, so as to solve the problem of large-scale skeleton tracking. In addition, the method has a certain ability to solve the occlusion and self-occlusion problems which often occur in human pose estimation. When a joint cannot be estimated accurately, its measurement is removed and it is tracked with the exact measurement from the other cameras. In the end, we use the Kalman consensus filter to track the human skeleton as a comparative experiment, indicating that the proposed method has a better performance.