Improvement-based lower approximation non-negative matrix antipode geometric estimation method

Abstract

The present invention relates to an improvement-based lower approximation non-negative matrix antipode geometric estimation method, relates to a computer vision technology, and provides the improvement-based lower approximation non-negative matrix antipode geometric estimation method along with outlier and model hypothesis pruning techniques. The method comprises the steps of firstly using a mismatching pruning technology for analyzing the relation between the matching pairs and eliminating the influence of the outliers (the mismatching points); then selecting a meaningful model hypothesis byusing a model hypothesis pruning technique; and then introducing a spatial constraint item (the spatially adjacent data points are more likely to belong to the same model hypothesis) and a sparse constraint item (the sparse non-negative elements can better reflect the consistent characteristics of the data points on the model) to a lower approximation non-negative matrix; and finally, solving theu and v of the lower approximation non-negative matrix by using an alternating iteration method, and extracting a multi-structure model from u.

Description

technical field [0001] The invention relates to computer vision technology, in particular to a method for estimating epipolar geometry based on an improved lower approximation non-negative matrix. Background technique [0002] With the advancement of technology, computer vision has been integrated into all aspects of human life. The research task of computer vision is to efficiently convert natural scene information into digital information by simulating human visual function. The epipolar geometry estimation is to reduce the pollution of the real data by noise and outliers by analyzing the constraint relationship between the two images, which greatly improves the robustness of the model fitting. Model fitting is a basic research task in computer vision, and it has a wide range of applications in image stitching, 3D reconstruction, motion segmentation, plane detection, and vanishing point detection. [0003] For multi-structure model fitting problems, some decomposition-ba...

AI Technical Summary

Problems solved by technology

However, the multi-Gaussian sampling method it uses is very time-consuming, especially when the data contains a large number of outliers

Images