Computational Vision for Interaction with People and Robots

Facilities for sensing and modification of the environment is crucial to delivery of robotics facilities that can interact with humans and objects in the environment. Both for recognition of objects and interpretation of human activities (for instruction and avoidance) the by far most versatile sensory modality is computational vision. Use of vision for interpretation of human gestures and for manipulation of objects is outlined in this paper. It is here described how combination of multiple visual cues can be used to achieve robustness and the tradeoff between models and cue integration is illustrated. The described vision competences are demonstrated in the context of an intelligent service robot that operates in a regular domestic setting. Keywords— Gesture Interpretation, Manipulation, Computational Vision, Pattern recognition, Cue Integration, Robotics