Stereo Matching And Reconstruction Of Three-dimensional Free Curve

At present,computer vision measurement has been used widely in industrial production,such as unmanned vehicle,defect detection,VR and so on.Especially,it has a great advantage on the measurement of some objectives with irregular geometric shape.Computer vision measurement means that using computer vision technology measures spatial geometry.With the advantages of simple structure,non-contact processing and high efficiency,it is widely used.However,this technology still exists many problems,such as,the low collection efficiency and image resolution,low precision measurement of low texture surface,description difficulty of irregular surface.Around the above problems,the main work of this paper is listed as follows:First,the improved background difference search algorithm and integrity detection algorithm are proposed to solve the problem of low efficiency and image resolution caused by manual collection of images.Both algorithms realize the automation of image acquisition and obtain the clear boundary of the objectives.The foreground area is used as the evaluation window to keep image clearity and the evaluation function unimodality.Finally,an automatic vision measurement system is established.Secondly,an optimized window matching algorithm is proposed in view of the poor precision measurement of low texture surface.First of all,we adopt the meanshift image segmentation method based on the stereo matching framework of Tao.Secondly,we use adaptive weighted window matching algorithm to obtain the initial disparity map,and incorporate with the optimal window algorithm to get more matching points.Finally,the parallax model is constructed as well as parameter calculation and optimization.As a result,the dense disparity map is obtained according to the established model.The matching result is compared with the three classical matching algorithms.It shows the advantage of our method in low texture area matching.Finally,a reconstruction method based on line-to-surface is proposed to solve the description of irregular geometric shape.This method use a laser to project a curve on the object surface.Then,this paper reconstructs the projective curve in three dimensions.The curve is fitted by B-spline interpolation method and the fitting error is analyzed.At last,we reconstruct the entire surface by repeatting above reconstruction process.Meanwhile,some experiments are performed to verify the reconstruction correctness of the curve.The experiments show that this method can reduce the number of mismatching point,improve the measurement precision and describe the surface shape of the object effectively.It is concluded that the average error of the proposed algorithm is less than 1mm in the low texture area.