| A real-time and robust human tracking robot system is presented. The robot is able to follow a walking human target based on the information obtained from a colour video camera and a laser range finder, which are mounted on the robot. For robust tracking, a model for the target person, which includes the colour and edge orientation histograms from the image data and the location of the person from the laser data, is constructed at the beginning of the tracking task. People in the field of view of the robot, including the target person, are detected by a leg detection algorithm developed in this thesis using the laser data. The target to be tracked is verified by comparing the image information with the target model. A particle filter is used to estimate the position and the velocity of the target person for guiding the robot's movements. Ego-motion compensation is embedded in the particle filtering to account for the effect of the motion of the robot itself in order to improve the tracking accuracy.;Some key requirements for robust and real-time human tracking, such as computation time, occlusion-handling, and capability of adapting to different targets are discussed. |
