| Obstacle detection plays an important role in the research of navigation technology for mobile robot. In this article a new method was studyed and designed for robot obstacle detection, where an algorithm of binocular stereo vision-based planar projection transmission was used. Using this method, the image matching and the generation of panoramic depth of field was avoided. On the other hand, the algorithm was simple, reliable, and suitable especially for indoor or highway environment with relatively smooth and basically monocular floor. Finally, application result of this method to the mobile robot was given. It includes two parts:planar projection transformation and obstacle detection.In the part of planar projection transformation, we propose a new stereo method for 2D navigation without depth search and metric camera calibration. Conventionally, there is an effective stereo method based on the constraint that an observer moves on the ground plane, namely the GP constraint. Let one of two images is a reference image. Then we transform the other image to the reference view assuming that all image points arise from the ground plane. Comparing the reference image with the transferred one, the points on obstacle areas have some disparities due to the heights while those on the ground plane do not. Therefore, subtraction of these images enables us to extract obstacle regions that have non-zero difference and it is unnecessary to search for correspondence between stereo images. We apply the pseudo-projective camera model that provides a good approximation to the projective camera in road scenes.In the stereo obstacle detection part, two different algorithms for obstacle detection are presented in this paper each based on different assumptions. They return only yes/no answers regarding the presence of obstacles in the field of view; no 3D reconstruction is performed. They have the advantage of fast determination of the existence of obstacles in a scene based on the solvability of a linear system, The first algorithm uses information about the ground plane. While the second only assumes that the ground is planar. In theory, it is possible to separate obstacle areas from a free space using the determined GP constraint with a pair of lines on the road. In practice, however, it is inevitable to have erroneous estimation at points around which steep intensity gradients are observed such as near the boundaries of road textures. By Morphological filtering, we remove the small area ("parallax" small point), and get the obstacle regions.Experiments have shown that this method is correct and feasible, and it avoided the problems in image matching and easy to achieve, it also proved the relationship between the two images when two video cameras placed parallel. |
