Research On The Plane Fitting Technology Based On Binocular Stereo Vision

Computer vision is a research topic which involves multi-field,multi-disciplinary，and binocular stereo vision as a very important research branch ofcomputer vision technology, it has many excellent features, such as small size, lowpower consumption, low cost and high efficiency, etc, to simulate the human eyestructure. Therefore, the technology has become a hotspot of research in machinevision.The paper, processing two-dimensional information on two different images of anobject photographed by a video camera below the same space sce ne, and restoring thethree-dimensional geometric information of the object, in other words, getthree-dimensional point cloud plane.(Point cloud is the spatial distributioncharacteristics and the collection of a large number of points). Then the paper uses thetraditional least squares algorithm and the RANSAC algorithm proposed in this paperto do plane fitting, transform three-dimensional point cloud plane into space planeequation. This process is of great significance in the actual three-dimensionalmodeling.The background of this paper is the realization of actual scene object plane fittingby means of binocular stereo vision system. By focusing on obtaining images bybinocular stereo vision system, camera calibration, matching and optimization of th efeatures of the images, and plane fitting, etc. The paper carries out the followingresearch work:1. Obtaining the left and right images of objects in the real world scenarios by thecamera and getting the two-dimensional matching points by using SIFT featurematching algorithm. There are a large number of false matches in these matchingpoints, so the paper does optimization process using three matching optimizationalgorithm. After comparing the results of each experiment, the paper ultimately selectsthe robust improved RANSAC algorithm which combines with Newton iterationalgorithm as the solution to optimize the match processing.2. To obtain a three-dimensional point cloud plane, the experiments must be doneunder three-dimensional point reconstruction algorithm and change two-dimensionalpoint coordinates into three-dimensional coordinates after optimizing the match. Thenthe paper use the method of least squares and the method of improved RANSACalgorithm which combines with eigenvalue method to do plane fitting experiments, including simulation and the experiment in actual scene, both experimental resultsprove that the method of improved RANSAC algorithm which combines witheigenvalue method can be well adapted to the various errors and outliers, and stablygot a better plane parameter estimate, so the RANSAC algorithm is a robust planefitting algorithm.