Three-dimensional Reconstruction Based On Binocular Stereo Vision Theory

As an important symbol of the intelligent robot, Stereo vision technology isone of the most active branches of robotics research. Based on the binocularstereo vision theory, making three-dimensional reconstruction as the target,this paper analysed and compared all stages of the process to achieve theobjective approach,at the end,summed up a applicable method ofthree-dimensional reconstruction under laboratory conditions.The establishment of imaging relation of corresponding points of objectsurface and image, is particularly important, in stereo vision imaging, so thecalibration of camera parameters is the first problem to be solved. Through theanalysis of model of Cameras inside outside parameters and lens distortion,based on OpenCV library checkerboard calibration method, this paper offereda method of calibration of the camera internal parameters.The image preprocessing is the basis and prerequisite of binocular stereovision3D reconstruction. In the third chapter of this paper mainly analyzedgeneral characteristics of the image. Using a top-down approach to achievethe image area detection, through image grayness, image filtering topreprocess the original image and histogram equalization, with the help ofOpenCV, the expected image local was obtained by experiments, thus, thereliability of this method was confirmed.The edge feature is an important feature of stereo vision matching, as itdirectly affects the accuracy of matching. In the fourth chapter we studied theimage edge detection, especially analysed several operators, such as canny,Sobel, Roberts, Prewitt and LOG. The OpenCVprogram renderings were compared and the canny was proved that is thebest operator in the processing of image edge detection. It can accuratelyextract the contour of the edge of the object. Through the extracted edge,feature points were extracted at last.Stereo matching is the main difficulties of stereo vision3D reconstruction. After the analysis of stereo matching characteristics, this paper offerd a quickmatching method, whose high applicability was proved under laboratoryconditions. At the end, with the held of this method, a complete experiment ofbinocular stereo vision3D reconstruction was carried, and experimental resultsare given.