Research Of3D-measurement Technology By Binocular Vision

The rapid development of modern manufacturing produces the requirement of measuring the product quickly and accurately. Today, the development of the computer vision technology promote the wide range of applications of the computer vision measurement technology which has advantages of non-contact, on-line, low cost, high efficiency and high precision. According to the principle of stereo vision, we can obtain3D information of the tested objects through the2D images. The stereo vision technology has been widely used in robot visual navigation,3D automatic measurement of the products, diagnostic medical assistance and so on. Now, it is a hot research topic and has broad application prospects.The Binocular stereo vision technology is based on the principle of the3D imaging of human eye. The realization of the binocular stereo vision measurement technology is through taking the tested objects’pictures from different angles through a couple of cameras which have the same performance. And then to obtain3D coordinate information of the tested objects by using the technology of camera calibration, image rectification, feature extraction, stereo matching and parallax distance measurement. Finally, we can reconstruct the3D shape of tested objects.According to the basic theory of measurement based on binocular stereo vision, the paper introduces each of the key steps and algorithms of3D computer vision detection. Then propose a suitable algorithm by analyzing and comparing the specific algorithm. And then verifies the feasibility of algorithm by experiments.The main research work can be summarized as follows:(1) Research on the principle of cameras mathematical model and calibration, then completes images acquisition and camera calibration by programming. The Calibration data include the internal parameters which describe the camera’s internal imaging and the external parameters which describe the relationship of the camera with the external environment.(2) Analyzes several image preprocessing methods which can optimize the images and make the features more prominent, then completes the comparative experiments.(3) Research on the principle of projection correction, and then rectifies the calibration results. The rectified images can make the feature matching easier.(4) Focus on the stereo matching algorithm based on SIFT local feature descriptor which belongs to the sparse feature point matching, it is based on scale space and has good robustness on the illumination change, distortion and noise. The paper proposes improved methods to improve the matching accuracy and reduce the complexity of the SIFT algorithm. The introduction of the wavelet transform and the epipolar constraint, greatly narrowing the search scope and improving the speed of algorithm. Finally, we complete the programming.(5) After getting the corresponding match point, research on the method of restoring the spatial coordinates and obtaining the geometry information of the three-dimensional objects by using the principle of parallax distance measurement. Finally, we provide the example of3D measurement.