Luminosity insensitivity stereo matching method based on self-adapting Census conversion

Abstract

The invention discloses a luminosity insensitivity stereo matching method based on self-adapting Census conversion. Firstly, a self-adapting area based on a cross skeleton is determined according to the structure and color information of an image so as to obtain Census conversion windows of any shape and any size; secondarily, the hamming distance after Census conversion is used as a matching cost and a local optimization method is adopted to calculate an initial parallax; finally, a two-step extracting method based on parallax statistics column diagram and left-right consistency verifying is disclosed to organically integrate the self-adapting area of the cross skeleton into the extracting process to obtain a high-precision parallax diagram. The luminosity insensitivity stereo matching method based on self-adapting Census conversion can obtain the high-precision parallax diagram from a stereo image pair with differences in illumination intensity and exposure time, integrate the matching precision and robustness on amplitude distortion, and be better adapted to application context of visual navigation for unmanned planes.

Description

technical field [0001] The invention relates to a stereo matching method in a stereo vision system, belongs to the field of computer vision, and is used for obtaining high-precision dense parallax information under the condition that there are differences in illumination and exposure between left and right views, so as to provide three-dimensional depth information recovery based on stereo vision. Reliable guarantee. Background technique [0002] Stereo matching is an important task in the field of computer vision. It obtains dense disparity maps through binocular or multi-eye image matching, so as to perceive the 3D depth information in the scene. Many scholars at home and abroad have conducted in-depth research in this field. The current dense stereo matching algorithm can be divided into four steps to analyze and study: the calculation of matching cost, the accumulation of matching cost, the calculation / optimization of disparity, and the refinement of disparity. In thes...

AI Technical Summary

Problems solved by technology

However, the traditional Census transformation based on fixed windows has the problem of choosing the window size.

If the transformation window is too small, the signal-to-noise ratio will be too low, resulting in low discrimination of the matching cost, and it is easy to cause false matching to the low-texture area; if the transformation window is too large, too many outliers (Outliers) will be introduced, which will affect the matching precision

[0006] In addition, the research also shows that optimizing the matching cost based on Census transform by global method will lead to high computational cost; optimizing the matching cost based on Census transform by using local method based on fixed support window will make it difficult to obtain high-precision parallax images, making parallax Obvious foreground magnification in discontinuous areas

Images