Scene Analysis and Interpretation


A basic understanding of the structure and dynamics of the surrounding environment and its operational context is fundamental for the intelligent, cooperative behavior of a robot assistant. The robot must be able to analyse and interpret events in the surrounding world in order to react appropriately.

Scene analysis will be based on known 3D image analysis methods for visual sensors (cameras and laser scanners). These methods will be used to analyze and suitably represent the common work space. They must especially be able to follow all actions of the human partner. Safety aspects play a particularly important role here, both in the general sense of reliably detecting and tracking persons (e.g. for collision avoidance) and in the detection of especially critical situations, such as interaction with humans in the same workspace.
New research focuses both on the area of task-related interpretation of complex scenes (making it possible to infer manipulation sequences from the relations of objects), and on motion analysis (especially human motion) and interpretation in the task context. Both areas form an important prerequisite for effective interaction between people and robot assistants.

Basic research topics in this field include:

	recognition of spatiotemporal relations
	static analysis of scenes and context
	identification of the human partner and recognition of his movements, gestures, and mimicry in the context of partner communication and system safety
	analysis of interaction patterns
	recognition of critical safety situations
	recognition of user intention
	execution of recognition tasks with the participation of the human partner (e.g. through feedback, inquiries, gestures, etc.)
	interactive learning of models of the enviroment