The authors propose an original approach that aims to treat aerial scenes which present a road traffic. They consider the scene as a succession of fixed images. Their system consists essentially of three parts: the segmentation of the acquired images, the modelization of a graph in which primitives provided by the segmentation are connecting according to perceptual organization and motion criteria, and the bi-partition of the graph by normalized cuts technique. The final aim consists in extracting vehicles from the background. Parameters of the proposed algorithm are chosen after a learning stage in which they maximize the similarity between manual cut and normalized cut results. Genetic algorithms are used in this step of optimization. Experimental results are very encouraging. This work opens very interesting perspectives. Firstly, improving the correspondence setting allows to ameliorate algorithm performances. Secondly, establishing a larger data base to refine learning and to have valid thresholds for a different sequences acquirement conditions. Finally, this work can serve us as a basis for a road traffic diagnosis or for implementation of a tracking system.
Segmentation and partition of aerial scenes for road traffic diagnosis
Segmentierung und Zerflegung von Luftbildaufnahmen für die Straßenverkehrsdiagnose
2004
6 Seiten, 8 Bilder, 1 Tabelle, 14 Quellen
Conference paper
English
Dynamic Segmentation of Traffic Scenes
British Library Conference Proceedings | 1995
|Fast Multi Model Motion Segmentation on Road Scenes
IEEE | 2018
|Color-based road detection in urban traffic scenes
IEEE | 2004
|Color-Based Road Detection in Urban Traffic Scenes
British Library Conference Proceedings | 2004
|Color based road detection in urban traffic scenes
IEEE | 2003
|