Observing and Interpreting Complex Human Activities in Everyday Environments
Jan Bandouch · 2011
In this thesis we developed a system for observing and interpreting complex human activities to enable novel interactive application scenarios e.g. in robotics. We describe the design, implementation and empirical evaluation of unintrusive methods for human fullbody motion capture. Our system is accurate and reliable even in the case of occlusions, object manipulations or interactions with the environment or another person. This is achieved by combining an anthropometric human model with novel hierarchical sampling strategies for probabilistic state estimation and implicit environment models. In addition, observed activities can be recognized using learned and self-adaptive motion models.