Multi-person vision-based head detector for markerless human motion capture
Charence Wong, Zhiqiang Zhang, Stephen McKeague, Guang‐Zhong Yang · 2013
Pervasive human motion capture in the workplace facilitates detailed analysis of the actions of individual subjects and team interaction. It is also important for ergonomic studies for assessing instrument design and workflow analysis. However, a busy, dynamic, team-based environment, such as the operating theatre poses a number of challenges for the currently used marker-based and sensor-based motion capture systems. Occlusions and sensor drift can affect the accuracy of the estimated motion. In this paper, we present a motion capture system that uses a vision-based head detection algorithm and a markerless inertial motion capture for estimating the motion of multiple people. The pose estimation obtained through inertial sensors is combined with location obtained through vision-based tracking to reconstruct the motion of each subject. A multi-target Kalman filter is used to track the movement of each subject. To handle the close proximity of the subjects, visual features associated with the body are used for data association. Experimental results demonstrate the accuracy of the proposed system.