Detecting collision-free paths by observing walking people

This paper presents a vision-based method for mobile robots to detect collision-free paths by observing walking people. The method extracts person´s regions from each image frame by background subtraction, and recovers paths from the trajectories of persons in successive frames. Due to the uncertainties inherent in the processes, many possible paths may be obtained. To cope with the problem, the obtained paths are classified according to their reliabilities, and are recorded on a set of maps with different levels of reliabilities, called hierarchical reliability maps (HRM). We describe the methods for estimating uncertainty in the image processing, quickly updating HRM, as well as robust path planning by using HRM.