Chiang Mai Municipality (Thailand) has a crossroads video surveillance, which is used for controlling the traffic, to trigger an alarm in case of emergency situations, like accidents. Image sequences are centralized at the Area Traffic Control (ATC) that requires continuous human supervision, in order to take essential decisions. The needs for an automated system for processing masses of information is highly required. The aim of this thesis is to investigate the automation of image and video analysis, to reduce the penibility of human surveillance, and assist supervisors of the road traffic safety in their decision making.
Chiang Mai Municipality has provided us a large volume of video sequences acquired under various conditions : daytime and night-time sequences, dry and rainy weather, quiet traffic and intensive traffic scenes, presence of various objects (pedestrians, cyclists, cars, vehicles).