| Binocular vision is an important branch of computer vision, by two cameras at different locations or one camera move or rotate to shoot for the same scene, get left and right two images. Then, computing spatial point about two parallax images to obtain three-dimensional coordinates of spatial points. Compared with other methods to obtain three-dimensional information, the way of binocular stereo vision by simulating human eyes to deal with the scene has the advantage of simple and reliable. In many areas have considerable value, Industrial parts such as three-dimensional measurement, robot navigation, virtual reality and military fields as well. In short binocular stereo vision is an issue with very significance and application prospects.This paper introduced the basic principle of binocular vision, expounds the camera calibration, dimensional calibration, stereo matching,3D Information Acquisition and other important measurement techniques. Focuses on the stereo matching algorithm based on sift operator, to some extent, the algorithm improve the correct of the match, but is not very satisfactory in matching speed and accuracy. In this paper, the algorithm is improved, and by experimental analysis of three-dimensional measurement results, further validate the feasibility of the algorithm. The main contents are as follows:1. Study the mathematical models and camera calibration principle, traditional calibration methods, use Zhang zheng you calibration method to get around both the internal and external parameters of the camera. Using the principle of binocular vision epipolar geometry to get two images dimensional correction, allows two cameras to achieve the desired level of visual binocular stereo structure.2. Research the stereo matching algorithm based on SIFT operator, a method of extracting feature points, generate feature vector, removing mismatching points. Based on this algorithm to make a improvement, propose a method to simplify SIFT operator, from the original 128 dimension reduced to 24 dimension, reducing the number of image matching feature points. Combined epipolar geometry, to get the matching imagestereo correction. Make the match search range from 2D down to one-dimensional, shorten the search time. After the first time match.combine with RANSAC algorithm, exclude the mismatching points, leave the good match,to further improve the matching accuracy.3. Build stereo vision three-dimensional measurement experiment platform, combine with OPENCV library, programming to reliaze the software system, by collecting images, three-dimensional calibration, stereo image correction, matching steps, finally, get a three-dimensional coordinate information. Experimental results show the system has better detection accuracy. |
