Abstract
This paper describes a method for detecting persons walking in a passage way from two consecutive TV images at long intervals. The method consists of three steps: 1) extracting regions of objects in a passage way from each TV image, based on the mean and the variance of brightness, 2) finding persons in the regions of objects on the basis of their shape and size and locating the persons in a passage way, 3) finding the correspondence of the persons between two consecutive images on the basis of their locations. The movements of persons are obtained by locating them in each image. The range information about persons is calculated from the decline of a TV camera, its height and other parameters. The method is effective even if the size and shape of walking persons change to a great extent and, even if the number of persons are not always equal between the images. There are many applications for this method. One will be vision systems for autonomous vehicles which can move around on a passage way for guiding the blind. In order to catch up with the movements of persons, the processing for finding a path should be done fast, say every 0.7sec. We used special hardware called DIP for the high speed processing.
