Image stereo matching method

Abstract

The invention belongs to the technical field of image processing, and particularly relates to an image stereo matching method. An existing stereo matching method based on deep learning is poor in generalization ability, the main reason is that data in the stereo matching field has certain defects such as too strong reflection and shielding, and on the other hand, an area lacking textures also exists in an image, so that some methods for adding neighborhood constraints are prone to over-fitting. Meanwhile, a high-quality data set in the stereo matching field is lacked, and a network with generalization ability is difficult to obtain through simple training. The invention provides the image stereo matching method which comprises the following steps: constructing a training image library; enhancing a training image; extracting point (line) features of all images; training a feature point extraction network; extracting unary features of the binocular image; acquiring a coarse precision disparity map by the iteration module; performing unitary feature aggregation; performing parallax regression; and performing disparity refinement. The problem of overfitting in a stereo depth matching algorithm is solved.

Description

technical field [0001] The present application belongs to the technical field of image processing, and in particular relates to an image stereo matching method. Background technique [0002] Stereo matching can be applied in scenarios such as autonomous driving and 3D reconstruction. In the stereo matching task, the occlusion and reflection problems and the lack of texture will have a greater impact on the matching results (such as figure 1 , there is reflection on the car glass in the outdoor road scene). In an ideal situation, the corrected binocular images remain consistent in the vertical direction, and there is only parallax in the horizontal direction. However, in actual situations, the correction of the camera is not perfect, and the parameters of the two cameras are not exactly the same. On the other hand, the scenes seen from different perspectives are not completely the same, and the scene content and lighting conditions will be different. , which will cause som...

AI Technical Summary

Problems solved by technology

[0004] The generalization ability of the existing stereo matching method based on deep learning is relatively poor. The main reason is that the data in the field of stereo matching has certain defects, such as too strong reflection, occlusion, etc. On the other hand, there will also be areas lacking texture in the image. This makes some methods that add neighborhood constraints prone to overfitting

At the same time, there is a lack of high-quality data sets in the field of stereo matching, and it is difficult to obtain a network with generalization ability through simple training.

Images