Dense stereo vision is a key technology for natural scene understanding. Recent progress in real-time dense stereo provides high quality depth information for (almost) every pixel of an image. Based on this information the traffic environment can be modelled precisely and completely. This includes a vertical modelling of the driver corridor, the determination of the free space in front of the car and the representation of all vertical obstacles delimiting the free space. So called stixels will be proposed to represent the 3D scene efficiently. Distance and height of each stixel are determined by the parts of obstacle it proximates. The stixel representation is designed to act as the common basis for the scene understanding tasks of driver assistance and autonomous systems. It will be shown that the inherent spatial integration delivers depth information with unprecedented accuracy.
Robust and precise 3D-modeling of traffic scenes based on dense stereo vision
Robuste und genaue 3D-Modellierung von Verkehrsszenen auf der Grundlage von dichten Stereoabbildungen
2009
9 Seiten, 8 Bilder, 6 Quellen
Conference paper
English
Stereo based visual odometry in difficult traffic scenes
IEEE | 2012
|Stereo Based Visual Odometry in Difficult Traffic Scenes
British Library Conference Proceedings | 2012
|Temporal analysis of stereo image sequences of traffic scenes
Automotive engineering | 1991
|Temporal Analysis of Stereo Image Sequences of Traffic Scenes
SAE Technical Papers | 1991
|