Related entropy-based Torr-M-Estimators basic matrix robust estimation method

Abstract

The invention discloses a related entropy-based Torr-M-Estimators basic matrix robust estimation method. A Torr-M-Estimators algorithm performs weighting processing on a residual error of each matching pair to reduce the influence of an exterior point of the relatively large residual error on an estimation process of a basic matrix, and different weight functions have different optimization results; an information theory-based related entropy function has relatively great advantage in robustness and is more remarkably improved in precision and robustness; data normalization is performed by utilizing a Hartley method, an initial value of the basic matrix is obtained by a linear least square method, a related entropy is a Torr-M-Estimators weight function, a rank of the basic matrix is limited to be 2 by eigenvalue decomposition, and a loop stop condition is determined by a minimum iterative error and a maximum iterative frequency; and an experiment shows that the method has very high estimation precision and robustness and is an important improved method in the basic matrix solving step in the fields of three-dimensional reconstruction, motion estimation, matching tracking and the like.

Description

technical field [0001] The invention belongs to the field of multi-view geometry in computer vision, and is a research invention for solving the basic matrix of the mathematical expression of epipolar constraints in stereo vision, and specifically relates to a Torr-M-Estimators basic matrix robust estimation method based on correlation entropy. Background technique [0002] In the field of computer vision, multi-view geometry is an important basis for computers to express the three-dimensional world through two-dimensional images. figure 2 As shown, the corresponding point m' of the image point m on another image plane must be above its epipolar line e'. The epipolar geometric relationship can be expressed by a matrix of order 3 and rank 2, which is called the fundamental matrix, also known as the F matrix (Fundamental Matrix), and the mathematical expression of the epipolar geometric relationship is m' T Fm=0, where m', m is the matching pair of corresponding feature point...

AI Technical Summary

Problems solved by technology

Based on this problem, people have done a lot of research. The solution methods are divided into three categories: linear method, iterative method and robust solution method. The linear solution method is fast, but when there are wrong matching points, the accuracy is very low. ; The iterative solution method has high precision, but it takes a long time to calculate compared with the linear solution method, and the result is still not very good when there are many error points; robust solution methods such as the minimum median method (LMedS), random sampling consistency method ( RANSAC) and its modified MLESAC, maximum a posteriori consensus algorithm (MAPSAC), M estimation method (Torr-M-Estimators), etc., these methods have a better effect on removing wrong matching points and bad points

Images