Research On Stereo Matching Technology And Its Hardware Realization

Binocular stereo vision technique, which directly simulates the physiological structure of human binocular vision, is an important technical means in the depth information extraction of three-dimensional scene. In recent years, with the research on stereo matching algorithm and the development of FPGA, ASIC and other hardware technologies, it can be widely used in robot navigation, UAV positioning and three-dimensional measurement. The key of the binocular stereo vision technology lies in stereo matching. In this paper, we have done a lot of research on how to improve the accuracy of stereo matching and how to realize real-time calculating in dense disparity map on hardware. The main contents are as follows:A new matching algorithm that based on the color in related and adaptive weighting is proposed on account of the traditional method of adaptive support weight (ASW). The method based on the user-define of inter color correlation can effectively reflect the matching small differences between pixel, it combines the color similarity, Euclidean distance similarity and inter color correlation similarity to distribute the weight value of pixels in the matching window so that it determines the left and right images matching cost relationship. Meanwhile, matching pixel are processed by improved rank transform, and then we calculate the cost aggregation to prevent the brightness and noise tampering with the matching results. Experiment results show that our algorithm can obtain better in stereo matching than traditional method of ASW.In the study of stereo matching algorithm based on feature points, extraction and location of image features is the key of feature matching, so we propose a new corner detection operator using the red and blue discs as a drone. The operation of color separation for the red and blue discs in order to get a gray image which has strong contrast and obvious symmetry. According to the new symmetric operator compute the response value of each pixel in the image, acquire a unique and strong robustness feature corner. Experiment results of Matlab simulation and hardware implementation based on FPGA platform show that proposed method is easy to implement in hardware technology, has higher accuracy, good anti-noise and faster of feature point extraction.For the binocular stereo vision have some disadvantages, such as large amount of computation and the poor real-time performance on the personal computer, so we propose a real-time hardware system for calculating dense disparity map based on FPGA. The entire stereo vision system include binocular video captured, rectification, stereo matching, disparity calculation, refinement and VGA display. According parallel pipeline technology in FPGA and integral image, proposed two new methods to achieve rapid SAD and simplified ASW stereo matching. Finally, test simulation of the real scene, to prove that the method can reduce the FPGA hardware resource and acquire real-time, accurate dense disparity results.