Study On 3D Reconstruction Of Free-form Surface Based On Binocular Vision

Along with the rapid development of science and technology, it appears more and more products with free-form surface. Because the free-form surface is hard to be described precisely, it becomes a technical problem that how to reconstruct the free-form surface to measure the data of surface. Binocular vision 3D reconstruction is one of the main method to realize the 3D reconstruction. In this paper, some key technologies of binocular vision 3D reconstruction based on grid projection are intensely researched aiming on free-form surface. The main content include below items.1. Considering the surface texture and unpredictable curvature, the 3D reconstruction method based on grid projection is proposed. The texture of free-form surface can be easily identified under grid projection. According to the requirement of the method, appropriate cameras, lens, illumination are chosen to set up the 3D reconstruction system of free-form surface based on grid projection.2. In the preprocess of images, a thinning method based on curve fitting aiming to optimize the distortion in the intersection of border lines and internal lines which influence accuracy of reconstruction is proposed. The method thins lines precisely on the basis of lines intersection and cuts down the reconstruction error of 0.2mm in some parts of surface.3. Image matching is the key step of binocular vision 3D reconstruction. A method combining the feature-based with area-based matching algorithm is proposed. The method realizes the matching according to two steps. Firstly, feature points belong to the lines and intersection are extracted and matched based on features. Secondly, an adaptive window Census transform matching method is proposed to matching points between lines which realizes precisely matching of binocular images.4. For the problem of quantitative error analysis of 3D reconstruction models, this paper compares the reconstruction models to standard models to analyse the accuracy of reconstruction method. The quantitative error analysis includes standard error, error of every points and error distribution on the surface.Experimental results show that the binocular vision 3D reconstruction method of this paper can effectively reconstruct several types of subjects and the error is less than 0.4mm.