Research On 3D Shape Measurement Technology Based On Computer Vision

In the information age with the rapid development of science and technology,three-dimensional information acquisition and processing technology is an important symbol of the development of machine vision and artificial intelligence.Numerous scientific researchers have been working on how to accurately obtain three-dimensional information of the objective world.Three-dimensional topography measurement technology can transform three-dimensional information into digital signal,which provides a convenient and feasible way for the three-dimensional digitalization of objects.Therefore,it is widely used in many fields such as industrial production,reverse design,cultural relics protection and so on.In this paper,two kinds of three-dimensional morphology measurement methods,structured light and visual Simultaneous Localization and Mapping(SLAM),are studied.Firstly,a three-dimensional topography measurement system based on monocular structured light is designed,and the point cloud denoising is carried out by using the three-dimensional point cloud fusion denoising algorithm.Then a three-dimensional shape measurement system based on monocular vision SLAM is designed,and the improved ORB algorithm is used for key frame feature matching.The main contents of this paper are as follows:(1)A three-dimensional topography measurement system based on monocular structured light is designed,and the denoising algorithm of three-dimensional point cloud fusion is used to denoise the image.Firstly,the hardware and software of the three-dimensional topography measurement system are designed,and the three-dimensional measurement is carried out by combining Gray code and four-step phase-shift method.Then,using the geometric invariance of marker space and coordinate normalization calculation,the cumulative error of point cloud stitching is reduced,and the registration of three-dimensional point cloud data is realized.Finally,the point cloud is denoised by the fusion point cloud denoising algorithm in this paper.The weighted fuzzy c-means denoising algorithm is used to remove the large-scale point cloud noise,and the fast bilateral filtering algorithm is used to filter and denoise the other small-scale point cloud noise to achieve the three-dimensional topography measurement of the target object.The experimental results show that the system has high three-dimensional measurement accuracy,small errors,good denoising effect of three-dimensional point cloud,and strong practicability.(2)An image feature matching method based on improved ORB algorithm is proposed.Firstly,the FAST feature extraction algorithm is improved to construct a non-linear scale space.The image pyramid is constructed by non-linear diffusion filtering method,and the direction of feature points is calculated by gray centroid method.Then the sub-sampling mode of FREAK feature description is optimized to remove the redundant information in the sampling points,and the improved descriptor is used to construct the feature vector.Finally,GMS algorithm is used to eliminate the false matching point pairs and reduce the occurrence probability of false matching point pairs.The experimental results show that the improved ORB matching algorithm has good real-time performance,high matching rate and strong robustness.(3)A three-dimensional shape measurement system based on monocular vision SLAM is designed.Based on the ORB-SLAM algorithm framework,the improved ORB feature matching algorithm is used to extract and register the features of the image frames collected by the camera.Then,according to the registration points of the improved ORB algorithm,the position and pose of the camera are estimated with the polar constraints,and the L-M algorithm is used to reduce the cumulative error,so as to realize the non-linear optimization of the position and pose of the camera.Then,the depth estimation is carried out by using triangulation algorithm to complete the transformation of three-dimensional point cloud.Finally,three-dimensional point cloud registration and fusion algorithm are used to measure the three-dimensional topography of the target object.The experimental results show that the quality of the three-dimensional point cloud data measured by this system is high,and the effect of three-dimensional topography measurement is good,which has certain practical value.