Binocular parallax estimation method introducing attention map

Abstract

The invention discloses a binocular parallax estimation method introducing an attention map, and particularly relates to a method for obtaining global information, generating an attention map and guiding binocular parallax estimation by using deep learning data learning capability. According to the attention map provided by the invention, the global features and semantic structures of an image arebetter extracted by leading out independent branches, the obtained attention map acts on the cost in a weighted mode to play a role in matching and guiding, and it is ensured that regions with the same semantic structure have parallax in smooth distribution. Meanwhile, the invention provides a strategy for fine adjustment based on sparse labels. Different supervision strategies are adopted in different fine adjustment stages, the optimal effect of the method can be achieved on a sparse data set through reconstruction error guidance, sparse correction and smooth constraint optimization, and the problem that the effect of a label-free area is poor is solved.

Description

technical field [0001] The invention belongs to the fields of image processing and computer vision, and relates to a method for estimating binocular disparity by introducing an attention map, in particular to a method for obtaining global information by using deep learning data learning ability, generating an attention map, and guiding the estimation of binocular disparity . Background technique [0002] The binocular depth estimation is to obtain the corresponding disparity value according to the relative position of each pixel between different views through two calibrated left and right views, and restore the disparity to the depth information of the image according to the camera imaging model. The existing binocular depth estimation methods are mainly divided into traditional methods and deep learning methods. [0003] The traditional method is divided into local algorithm and global algorithm. The local algorithm uses the similarity of the neighboring pixels in the win...

AI Technical Summary

Problems solved by technology

[0005] The present invention aims to overcome the deficiencies of the existing deep learning methods. The present invention proposes a new idea to solve the problem of mismatching large areas without texture, that is, the attention map; by drawing independent branches in the binocular parallax estimation network, using In order to better obtain global information and semantic structure information, the obtained attention map acts on the cost through weighting, and uses the obtained semantic structure information to guide the binocular stereo matching, correct the mismatching area, and ensure the same semantic structure Regions have a smooth distribution of disparity

At the same time, the present invention proposes a fine-tuning strategy on sparse label data, which solves the problem that the effect of unlabeled areas is poor and the fine-tuning effect is difficult to achieve optimal

Images