Research Of Three-Dimensional Reconstruction Technology Based On Binocular Stereo Vision

Computer Vision is a computer realization of human visual function, and the perception, recognition and understanding of three-dimensional (3D) scene in the world. Automatic character extraction from the image and reconstruction of3D objects is not only the issue concerned in the field of measurement, but also a live topic in the field of Computer Vision and Image Processing. Since3D reconstruction of binocular stereo vision technology shows high efficiency, low cost, portability, and great usefulness for achieving automation in the workplace, it has been one of the most important directions of development in the field of3D modeling.Based on the experimental system of binocular stereo vision, a detailed research has been done in this paper, which is concerned about image acquisition, camera calibration, image pre-processing, feature extraction, stereo matching and3D reconstruction technologies. In the end, the model of3D object reconstruction has been built successfully. The following text will introduce the work of this paper.Firstly, we analyze the basic measuring principle of binocular stereo vision, and design the experiment process and system architecture. Then we build an appropriate calibration template and use a secondary calibration method to complete the calibration experiment with Jean-Yves Bouguet Camera Calibration Toolbox. The internal and external parameters of the binocular cameras will be achieved after computer calculation.Secondly, according to the deficiency of pretreatment on the pair of traditional stereo images, a method of4-step image preprocessing is used to effectively remove image noise, which can also balance the difference of brightness and enhanced the edge and details in the images.Thirdly, we introduce three classical algorithms of feature-point extraction based on grayscale changing, including SUSAN, Moravec and Harris algorithms. With the consideration of the key functions, such as the ability of corner extracting, noise immunity, the success rate of matching, we make the comparison and analysis of the three experimental outcomes, evaluate and summarize their performances.Finally, the principle and characteristic of SIFT algorithm is introduced in details. After analyzing the performance of Harris corner extraction algorithm and SIFT feature matching algorithm, we decide to combine Harris with SIFT in an effective way. By the use of reconstruction algorithm of3-D spatial points, we complete the surface fitting of the resulting points and get the three-dimensional graphics. After carrying out a systematic analysis of the experimental process and results, we achieve the desired effect and make the overall conclusion of the entire paper.