Motion tracking for human-robot interaction

This thesis aims at improving the perceptional side of interaction to enable the handling of more complex situations. For the Swiss National Exposition Expo.02 the robot RoboX and an event-driven interaction system have been designed by members of our lab. Experiences in this exhibition underline the importance of detecting and tracking persons and the understanding of dynamic contexts for interactive systems. To understand dynamic situations, three types of objects are distinguished: static, dynamic and semi-static. Motion detection is used to identify dynamic objects and to separate the static environment from moving persons. Scan alignment corrects the position error of the robot. Thus, accumulating information over scans eventually results in maps of the environment. Dedicated object detectors identify the state and location of semi-static objects. Based on the motion detected a multi-object tracking algorithm collects information on persons in the vicinity of the robot. As an example, how this newly obtained information can be used in interaction, an automatic narration is generated for actions of visitors at Expo.02 using this system.