Reconstruction Method Of 3D Model Based On SURF And Sequence Images

Image-based reconstruction of 3D model is a hot and difficult research topic of computer vision field. Feature points detection and matching of image-based 3D modeling process are the key technology and they are largely affected the efficiency and accuracy of reconstruction results. This paper studies as follows:(1) 3D Reconstruction theory based on multiple images. This paper studies the 3D reconstruction steps including feature point detection and matching, camera self calibration (including obtain basic matrix, essential matrix, compute internal and external parameters of camera and projection matrix), and gets the coordinates of 3D space. This paper mainly focus on feature point detection and matching.(2) SURF feature extraction and matching algorithm. Feature points detection and matching is the base of 3D reconstruction and the performance of feature point's detection algorithm is very important. This paper studies SURF algorithm and carries on experiments to analyze its performance. Compare with SIFT algorithm in feature point detection's precision, feature points'number, and speed three aspects. Verify the superiority of SURF in this paper.(3) Dense matching algorithm of feature point. Although the SURF algorithm obtains the optimal performance in speed and accuracy, but it gets relatively small number feature points. To solve this problem, apply the region growing matching algorithm to the feature points matching. Use having matched points for the seed points, expanding to around for getting dense matching points.(4) Realize a 3D reconstruction system of multiple images based on the above theories. Carry on experiments using photos and apply both SURF algorithm and region matching algorithm to the system. Compared with the system which uses SIFT algorithm only, the precision of reconstruction results improved a lot.