Research On Key Technology Of Binocular Stereovision Ranging

Binocular stereovision is an important research branch of computer vision, whichcan restore the three-dimensional depth information of the scene by computing thedisparity of space points in two images of the same scene. Compared with thetraditional depth measurement, it has two outstanding advantages, non-contact andpassiveness and it has been widely used in industrial inspection, aerospace, roboticsnavigation, battlefield surveillance and other fields with these excellent features.In recent years, binocular stereovision method has got greater progress. However,there are still many problems to be solved, such as camera calibration accuracy andstereo matching accuracy problems, which need the researchers to propose newsolutions continuously. The paper carried out theory and technical implementationresearch on the camera calibration, stereo matching method, and stereovisionmeasurement error and accuracy analysis for the binocular stereovision. The mainwork and innovations are as follows:1. In order to improve the accuracy of camera calibration, the corner extractionalgorithm based on linear fitting of the initial corner’s limited neighborhood isproposed. Harris operator is employed to extract the checkerboard cornerspreliminarily, the extracted checkerboard edge is implemented the linear fitting withinthe certain scope of the initial harris corners and the intersection of the two fittinglines is the final corner. The extracted corners are used to calibrate the stereo visionsystem with plane calibration method, which get higher calibration accuracy, thecorner extraction method is proved to be effective. 2. In order to get a better balance both in time and accuracy of local stereomatching algorithm. A fast stereo matching algorithm based on seeds pointpropagation is presented. The joint matching cost is employed to measure thedifference between left and right image in the algorithm, use the image edgeinformation to help build the dynamic matching window, which can weaken theadverse effects of the fixed window and obtain high quality initial disparity map.After the seeds point selection through the left and right consistency check andregional characteristics, the seeds points are propagated with a variety of constraints.At last, the disparity map is refined with the regional vote. The algorithm can get ahigher accuracy on most region of test images and has certain practical value.3. The semi-global stereo matching algorithm based on the disparity planarconstraints is put forward according to incomplete of the optimal path in thesemi-global algorithm. We establish sparse correspondences by extracting FASTfeature points and ORB descriptor in both images. The sparse disparity points areclustered with the image segmentation and propagate to other points in the samesegmented region. The reliable disparity points of the segmented region are carriedout the planar fitting and the difference between the plane fitting disparity and theinitial disparity is added to the pixel matching cost as a constraint. Then, the optimalsolution to the energy equation is obtained with the semi-global method. Experimentsshow that the algorithm can work well in the region with textureless and repetitivestructure and it can also get stable matching results in the occlusion and depthdiscontinuity region while maintaining efficiency.4. Some factors affecting the measurement accuracy in the binocularstereovision system are analyzed theoretically in detail. The theoretical derivation forthe three aspects of the camera pixel quantization error, structural parameters error,and range resolution accuracy are launched and the corresponding conclusions aredrawn. The two algorithms are tested in different scenes. At last, some referencedsuggestions for building the binocular stereovision ranging platforms are put forwardwith the method of combining theoretical derivation with the experimental dataanalysis.