Study On Dimension Measurement Based On Binocular Stereo Visual System

Accompanying the great advancement of computer version and machine vision, the binocular stereo visual system has been successfully applied to solve some practical problems. Research shows that the dimension measurement based on binocular stereo visual has wide perspectives, especially, in noncontact measurement. However, this measurement is unable to extract feature points accurately and the mismatching rate is high. On the basis of the research on these two problems, an improved SIFT(Scale Invariant Feature Transform) algorithm is put forward to fulfill the extraction of feature points and the epipolar constraint method is applied to reduce the mismatching rate. Finally, the proposed algorithm is applied to the design and development of the software and hardware platform. The main contents and results are summarized as follows:1.On the calibration of the binocular stereo vision system, the related coordinate systems are introduced and their relationships are given after analysis of the camera model. Then a camera model calibration algorithm based on the plane pattern is adopted to calibrate the binocular stereo vision and determine the corresponding relationship between the image plane and the three-dimensional scene plane as well as calculate the related parameters.2. Afterwards, a new algorithm is put forward to extract feature points after the introduction and comparison of the Harris Corner and SIFT Detector. Meanwhile, the binary searching algorithm is used to screen out required points. Finally, the experiments demonstrate the feasibility and reliability of the new algorithm.3. Furthermore, the SIFT matching algorithm is applied to fulfill the initial matching and the epipolar constraint is adopted to solve the problem of one-to-many mismatching. This rule reduces the searching scope from a whole image to a line. Specifically, the bidirectional matching is proposed to improve the accuracy.4. Moreover, we use the three-dimension reconstruction algorithm to obtain the spatial points according to the matching points and projector matrix achieved by system calibration. Then, the dimension can be deduced.Eventually, it is shown by the results of some experiments that compared with the conventional algorithms, the modified SIFT algorithm and epipolar constraint can achieve great improvement in extraction accuracy and match efficiency of the feature points, and the requirement of precision can be met if the distortion rectifying is adopted.