Visual Tracking of Multiple Humans with Machine Learning based Robustness Enhancement applied to Real-World Robotic Systems

und durch die Fakultät für Informatik am 29.08.2012 angenommen. This thesis presents a novel and robust vision-based 3D multiple human tracking system. It is capable of automatically identifying, labelling and tracking multiple humans in real-time even when they occlude each other. The primary contribution is a methodology to improve the robustness of the human tracking system and demonstrate its integration into real-world scenarios. The proposal is a system consisting of 2 stages, 1. a vision based human tracking system using multiple visual cues with a robust occlusion handling module, and 2. a machine learning based module for intelligent multi-modal fusion and self adapting the system towards drastic changes in lighting conditions. The function of the intelligent fusion module is to perform an on line analysis of image parameters that influence the performance of the tracker. According to this analysis, optimal weights are generated for each visual modality, determining its contribution in the current scene. The thesis also proposes a novel approach to validate the 3D multiple human track-ing system through zero-error ground truth data. Further, it proposes and demon-strates that the author’s work can be easily integrated into a variety of distributed robotic systems being used in real world applications. The main focus of this the-sis is in the area of Human-Robot Interaction, which requires real-time and precise information of the human positions to guarantee the safe interaction