Vision-based markerless three-dimensional human-arm tracking

Abstract

Vision-based human motion tracking has been used in surveillance, human-machine interfaces in robotics and automation, virtual reality applications and biomechanics. Most techniques require markers, use a predefined motion sequence or user-intervention for initialisation, and do not process in real-time. This thesis describes a vision-based non-invasive technique for markerless real-time tracking of human-arm motion. Human-arm motion is tracked by processing images from two calibrated cameras in real-time, to estimate the positions of the joint centres of the wrist and elbow in three dimensions (3D), and to compute the 3D positions of the index finger and thumb in order to permit determination of the orientation of the hand. The hand-arm motion tracking is implemented in a human-robot interface in performing tasks of picking up an object and placing it at a target location. Tracking of the hand and arm is carried out without any prior knowledge of subject's arm length, color, width and distance from the cameras.