ABSTRACT
Stereo vision provides the scene geometry independent from the scene appearance but is not discriminative enough in the far region due to the limited image resolution. In practice, typical stereo vision systems are costly because they should be carefully configured, i.e., the two cameras are parallel and displaced horizontally with respect to each other, so that corresponding image points are on the same row to reduce the searching region. In previous approaches, image appearance features and geometric features are fused for robust detection. The application of mobile robots, environment perception is crucial which mainly includes the detection of vehicles, pedestrians, traffic signs and drivable road regions. Drivable road detection is an essential functionality and has been widely studied in the past few years. Previous works generally use range sensors, vision sensors, and the fusion of multi-modal sensors. In terms of the physical configuration of vision-based road detection systems, monocular and binocular camera systems are commonly used.
