Image processing method based on improved AD-Census stereo matching algorithm

Abstract

The invention discloses an image processing method based on an improved AD-Census stereo matching algorithm. The method comprises the steps that a reference image and a target image are acquired, and the reference image and the target image need to meet epipolar constraints; determining a superpixel segmentation label map and an edge gradient map for obtaining the reference image; in the cost function calculation stage, in the label map and the gradient map, the number of pixels which are consistent with the label value of the center pixel and have the gradient value larger than a set threshold value in the matching window and the percentage of all pixels in the whole matching window are counted; taking the two obtained statistical magnitudes as independent variables of an improved AD-Census cost function, and adaptively setting the weights of AD and Census according to the form of an exponential function; in the cost aggregation stage, cost aggregation is carried out by adopting a mode which is the same as that of the original AD-Census stereo matching; and in the parallax optimization stage, parallax post-processing in the SGM algorithm is adopted to carry out parallax optimization. By means of the method, the stereo matching efficiency of the AD-Census algorithm in the smooth area and the target boundary area can be maximized.

Description

technical field [0001] The present application relates to the technical field of image processing, and in particular, to a method for generating a disparity map based on binocular stereo vision. Background technique [0002] It is well known that humans rely on the eyes to perceive the surrounding environment, and people can perceive the three-dimensional information of external objects through the eyes. The binocular stereo vision technology refers to the vision of the bionic human, and obtains the disparity map of the target object from the triangular geometric relationship of the target object in the left and right fields of view. Combined with the disparity map and through the transformation between coordinate systems in space, the three-dimensional information of the target object is finally obtained. . With the continuous advancement of science and technology, more and more artificial intelligence devices have entered people's lives, and most of the applications of th...

AI Technical Summary

Problems solved by technology

[0004] For the existing fixed weight AD-Census matching cost function, it may not be able to exert its maximum effect

Images