Détection des deux roues par capteurs vidéos fixes
Nicolas Tronson · HAL (Le Centre pour la Communication Scientifique Directe) · 2001
The objective of the work presented in this thesis is the creation of an optic system able to detect the two-wheeled vehicles in urban areas. To this prospect, we choose, owing to the limits of current systems primarily due to shadows cast by vehicles, thus penalizing their detection, to opt for a stereovision system. The originality of the system presented here is the optics and cameras placement choice. We use two cameras placed one above the other, oriented and aligned along a vertical axes. We choose this configuration for stability reasons, and because it makes it possible to cover the entire scene. Then, we study the system calibration, in one hand with a generic calibration tool for stereo systems, in the other hand with an approach specifically developed to consider the characteristics of the system. With regard to data processing, to perform the matching between images and thus obtain a 3D map, we propose three approaches. The first two are largely based on the state of the art. The third one, which was designed as a part of this thesis, is quite original and can target more precisely the search area by decomposing the 3D scene in different layers corresponding to different heights above the road level. To finally be able to detect, classify and track vehicles, 3D information providing the position and size of vehicles is integrated into an existing tracking software.