Falls on the track detection in public transport by artificial vision

摘要

The aim of this application is to improve safety by detecting automatically people or objects falling on or crossing the tracks in public transport. Through computer vision processng, the system to be developed determines and sends, in due time and via a satisfactory Human Machine Interface (HMI), alarms and the corresponding video image to the systems operators. In the current version of the system, to detect falls on the tracks, we propose to fix the cameras on teh ceiling, aimed vertically so that the framing delineates clearly what is standing on the platform from what is above the tracks. To detect falls, only the moving edges in the "tracks" part of the picture will trigger a warning. With this particular camera position, passengers bent over the edge are detected as possible incidents and potential accidents, though other camera placements do not detect such incident scenarios and are being considered with the French operator RATP. According to the very hard working conditions (-uncontrolled lightening, changes of background, unknown moving objects) a specific robust algorithm has been designed to perform the detection of the noving edges. This algorithm is based on the analysis of the differences of gray level between successive frames in an original video sequence of images. Edge extraction, performed on the difference of successive images, allows to retain only the edges of moving objects in the frame.