Research On Multi View Feature Points Extraction And Stereo Matching

The research and experimental which addresses feature points extraction and stereo matching based on the 3D reconstruction of multi-view images are described in this thesis.Feature extraction is the process to extract discernible characteristics from the images on the basis of computer vision algorithms.Stereo matching is the process matching surface points of three-dimensional objects to projected points generated by images acquired from different viewpoints.Feature extraction and stereo matching are important research topics in the field of 3D reconstruction and computer vision.The aim of this thesis is to extract the sub-pixel feature points which are beneficial for matching and evenly distributed,and get dense disparity robustness map.Research work are summarized as follows:(1)Due to the benefits of Harris corner extraction and SIFT descriptor,a feature point extraction algorithm based on Harris corner and SIFT descriptor is proposed.First,an improved Harris algorithm is used to extract the sub pixel feature points with uniform distribution.Then,the main direction is assigned to these feature points.Finally,the SIFT descriptor is established.The experimental results show that the algorithm extracts more accurate positioning feature points and preserve the invariance of rotation and change in illumination by using SIFT descriptor.(2)In order to improve the accuracy of feature matching,the thesis use the Euclidean distance ratio of the nearest neighbor and next nearest neighbor to obtain early matching pairs,then use two-way matching strategy and RANSAC algorithm remove mismatches.To address the problem of the selecting,an appropriate window size and decreasing redundant computation,a fast region matching approach based on disparity gradient is presented.This method judges the types of matching points and selects the appropriate window size according to disparity gradient and ascertains the search region by the previous point's disparity.The experiment result indicates that the proposed algorithm is the higher matching speed and theless calculation than the ones based on invariable search range.(3)According to advantages and disadvantages of both feature matching and region matching,a hybrid matching algorithm was proposed.First,use featurematching for feature points which extracted by improved Harris algorithm,then use the region matching as a feature point matching algorithm to match.At the stage of region matching,using canny edge feature extraction algorithm to further simplify the type judgment pixels and using feature points in window to further ascertain the search region.Experimental results show that the combined matching algorithm is feasible and the matching is more accurate in the edge region.(4)Finally,three-dimensional reconstruction of two viewpoint images were studied.On OpenCV + OpenGL development platform,we achieve camera calibration ? epipolar rectification ? feature extraction ? stereo matching,and use triangulation to compute the depth of standard image libraries and the measured images,the final complete 3D point cloud reconstruction.