The ability to perceive their surroundings is of vital importance for machines that naturally interact with humans. A challenging task hereby is the comprehensive recognition and understanding of human behavior. This involves simple actions, complex activities as well as facial expressions and body language. This research project is concerned with vision-based human motion capture and analysis. The main focus lies on the perception and description of the dynamics of human motion and the creation of general models for continuous movements. For the developed methods to serve a wide range of applications, real- world requirements are considered, such as robustness, operation in real-time, and the use of few camera views.
Research Objectives:
- Vision-based motion capture
- Dynamical motion models
- Action recognition
Detection and analysis of human activities for human-machine interaction
This project deals with the detection and analysis of human movements. A machine that is supposed to interact with its environment and especially with humans must be able to understand human movements and actions.
To do this, algorithms are first of all required that can detect and classify simple motion sequences in videos. From this, conclusions can then be drawn for more complex activities.
An important aspect here is the applicability under realistic conditions. This means that the system has to be real-time capable, work with dynamic backgrounds and partial occlusions as well as with variations of the viewing angle.
On a higher level is the semantic interpretation of the context, e.g. the recognition of intentions, behaviour patterns and the prediction of actions.