Very promising sensor systems are binocular stereo systems. In theory, stereo systems are able to fulfill many requirements, especially requirements about the field-of-view, the coverage range, and the capability to provide three-dimensional (3D) information of an observed scene directly by reconstructing using two synchronously acquired two-dimensional (2D) digital images. In realistic traffic scenarios, however, many problems concerning the reconstruction process have to be addressed. Examples are poor distance resolution for great distances, confusion due to correlation errors, and periodic patterns. All examples evoke ambiguous correspondence and, as a consequence, might lead to false 3D reconstruction. In order to be able to determine the quality of the reconstruction process we developed a set of tools. These tools have been tested successfully using a commercially available low-cost stereo system. As the scene reconstruction is used for obstacle detection and tracking, relevant quality features are the correctness and the resolution of the depth estimation. This work demonstrates the importance of a careful implementation of the stereo process. There are available stereo systems with a satisfactory implementation. However, from these observations it can be concluded that an elimination of ambiguous stereo correspondences should advance the state of the art in stereo vision.
Analysis of binocular stereo systems
Analyse von binokularen Stereosystemen
2008
5 Seiten, 8 Bilder, 3 Quellen
Conference paper
English
IEEE | 2008
|Online Contents | 2008
|Occlusions and Binocular Stereo
British Library Online Contents | 1995
|Binocular Stereo from Grey-Scale Images
British Library Online Contents | 1999
|Stereo fusion: Combining refractive and binocular disparity
British Library Online Contents | 2016
|