| As an important branch of computer vision, binocular stereo vision can be used to obtain the distance information of 3D scene. It mainly consists of six steps:camera calibration, image acquisition, image preprocessing, feature detection and extraction, stereo matching and obtaining distance information of 3D scene. This paper mainly focuses on the camera calibration, image acquisition and stereo matching.In the aspect of camera calibration, Zhang Zhengyou used 2D plane calibration method, combining with the calibration toolbox of Matlab, successfully realized the calibration of two cameras, respectively obtained the internal and external parameters of two cameras, and the rotation and translation parameters. The calibration result has very high precision, fully meets the binocular stereo vision system in this paper needs.Image acquisition is the basis of the binocular vision system. Improving the performance of image acquisition will improves the performance of the whole system. Based on the Zynq-7000 chip’s ARM and FPGA integrated feature, design and implementation the image acquisition module of binocular vision system. Through cameras’synchronous signals, image acquisition module on FPGA captures the images of two cameras, and through the chip’s internal bus, using DMA to transfer image data to DDR. This method improves the efficiency of image acquisition and transmission, reduces the burden of CPU, improves the performance of the whole binocular vision system.Stereo matching is one of the most important and critical steps in stereo vision. It uses regional feature, point feature, or phase feature, to match and then gets the result of the matched points. Based on triangle principle and matched points, the 3D points’s distance will be getted. This paper uses Harris+SIFT algorithm to relize stereo matching. This algorithm firstly uses the Harris method to detect corner points, and then uses SIFT method to get feature vector, at last, according to the SIFT feature vector, matching the fearure points of the two images. Using this algorithm is efficient, consumes less time, and gets better matching result than using the SIFT algorithm alone. |
