Design And Research On Binocular Stereo Vision Systerm For Robot Navigation

As technology advances, the intelligent level of robots is getting higher and higher. Robots are designed to assist or replace human to complete tasks. To achieve this target, robots should be able to fully perceive the relative position between robots and external environment. And visual measurement is a better choice. However, considering harsh conditions of momocular measurement and complexity of multicular measurement, this paper chooses binocular measurement. Binocular stereo vision could get the orientation and distance of objects in the scene. These information could help robots avoid obstacles and close targets, and then achieve the task of robot navigation.Binocular stereo vision technology includes three key parts, came ra calibration, stereo matching, reconstruction and so on.A mature planar checkerboard calibration algorithm has been used in this paper. For binocular vision system of the optical axises parallel to each other, this paper has presented an experimental method. The method has veritif ied the theoretical derivations, when only the baseline changes, there is invariance in inside parameters and mere variance in outside horizontal parameter. The application of the conclusion could reduce the calibration work.Images have been rectified before stereo matching. The method named combination of edge feature matching and sampling region matching has been proposed, which could be divided into the following steps. Firstly, extract edges of the color images, divide the target image into non-overlapping super-pixels, select MAD-Tanimoto similarity metric, match in the target image and then obtain the initial disparity map of edges. Secondly, correct the initial edge disparity, sample the non-edge region to match, and fill to obtain dense disparity map according to the disparity propagation mechanisms. At the same time, complete edges based region growing segmentation. If necessary, do Mean Shift smooth filter before segmentation. Last, correct the dense disparity to get the final dense disparity map according to the dense disparity and the result of segmentation.Based on the above stereo matching method in this paper, ensuring the effect of matching result and considering features of robot navigation, this paper has proposed two stereo matching methods based on sub-image. Compared with traditional full stereo matching method, these two methods have improved the time and space efficiency according to the results of experiments.According to the rectified inside-outside parameters of binocular stereo vision and the final dense disparity map, this paper has completed reconstruction of quasi-parallel binocular stereo vision and then got the orientation and distance of objects in the scene. Its realization is simple and convenient and t he reconstruction has been verified reasonable and accurate qualitatively and quantitatively.