Active vision-based moving objects detection from a Motion Grid

This paper proposes an active vision-based method for moving objects detection and tracking from a mobile robot. A Motion Grid, defined on the image, provides the probability to find a moving object on each image position; this data structure allows to fuse different approaches executed to detect and track moving objects. First a geometrical reasoning based on tracklets generates hypothesis; then a cognitive reasoning based both on model-based objects recognition and on contextual reasoning, confirms the presence of moving objects in the robot vicinity. Experimental results are presented using sequences acquired by a robot moving outdoor in a public space.