Research Of 3D Measurement Technology Based On Structured Light Stereo Vision

With the development of digital information, the needs of all walks of life to the three-dimensional measurement system is increased. Considering the structured light stereo vision system is non-contact, fast, flexible and portable, with good real-time performance and other advantages, a set of 3D measurement system is designed in this article based on structured light stereo vision for the measurement of large object which surface is smooth and has less texture feature, The main research work is as fllows.(1) Synthetically analyze the research situation of structured light 3D measurement technology at home and abroad. Design the research program of the 3D measurement system based on structured light stereo vision, and set up the experimental platform, complete the system calibra tion and analysis of error sources.(2) Study on different feature extraction algorithm, and the experimental results are compared with the experimental results. Select the Sojka algorithm and Harris binomal algorithm as the feature extraction method used by t he research, on the basis of this, the superiority of the pseudo random code in feature matching is verified. A stereo matching algorithm from coarse to fine is designed Combining the features of pseudo random code, similar metric and RANSAC algorithm. Meanwhile, the matching efficiency is improved by introducing the Pyramid model which reduces the matching search scope. Analysised by experiment, the NCC method is selected as the similarity measure ment.(3) Based on the KD-tree search strategy, mult groups of local point cloud data calculated by the matching feature points is registered and fused using ICP algorithm. As the result, a complete 3D point cloud model is completed. Due to the existence of overlapping regions and noise points of local point cloud data, the method of weighted down sampling is used to eliminate the redundancy of the registered point cloud, at the same time, the least square algorithm is used to smooth the noise. Finally, a new algorithm for triangulation is prop osed to reconstruct the 3D model from the smoothed point cloud, based on incremental growth. The reconstruction results of standard parts under two different kinds of feature extraction algorithms are compared with the measurement results of Go! Scan 3D scanner, and the accurac y of reconstruction is analyzed. After the experiment, the 3D model reconstruction and measurement can be well realized based on two kinds of extraction algorithm, Sojka algorithm and Harris binomial algorithm. In this article, the experimental measurement distance of about 185 cm, and the scanning range of 85.12cm?63.23 cm.