Automatic contour initialization for image tracking of multi-lane vehicles and motorcycles

The paper presents a novel design for automatic contour initialization and its applications to vision-based traffic monitoring. Active contour models are constructed subsequently to locate and track various types of vehicles, including cars and motorcycles. We propose to use image measurement to predict the position and size of vehicles in a special designed detection window. The initialization mechanism facilitates contour tracking of different size vehicle simultaneously at a junction. Actual CCD images of an artery junction were employed to verify the proposed image tracking system. Traffic parameters such as vehicle speed and numbers of vehicles are obtained with satisfactory accuracy.