The Study On Euclidean Reconstruction Of Three-dimensional Reconstruction Technology

Three-dimensional reconstruction is one core of computer vision research, which goal is recover three-dimensional information of space object through two-dimensional image information of the object,The main steps include feature point detection and matching, solution of the space constraint matrix,camera calibration,reconstruction of feature points,triangulation,texture mapping and so on. This article show some work about euclidean reconstruction that ultimate goal of this research field,which main content include:(1)This article first proposes an Harris improved algorithm,which can detect corner points and edge points at the same time to enhance the reconstruction results.(2)This article proposes an adaptive optimal grouping of feature point matching strategy in the aspect of matching. Following the histogram's properties,geting the optimal group of corner points in accordance with eigenvalue is the first step of the algorithm,then this algorithm use the Monogenic phase algorithm to match.This method can remove the error matching points effectively and reduce the feature point search time.(3)This article proposes an adaptive weighted iterative algorithm In three-view reconstruction,which is implemented by iterative screening of the weighted matching points, this algorithm can accurately solve the three-view matching points and further the trifocal tensor.(4) In the aspects of the hierarchical reconstruction,this article proposes an linear clustering detection algorithm which is based hough algorithm to get and classify the line information,Then combines with an improved ransac estimation algorithm to solve vanishing point,which improves the robustness of hierarchical reconstruction.(5) This article presents a reconstruction strategy by the the reference with significant features.This strategy avoids the interference of error feature points, and is usually used in reconstruction in complex cases and receives a satisfactory result.