A Fast Stereo Matching Network with Multi-Cross Attention

Ming Wei,Jiarong Wang,Jiaqi Sun,Yi Wu,Ming Zhu,Changji Liu

doi:10.3390/s21186016

Abstract

Stereo matching networks based on deep learning are widely developed and can obtain excellent disparity estimation. We present a new end-to-end fast deep learning stereo matching network in this work that aims to determine the corresponding disparity from two stereo image pairs. We extract the characteristics of the low-resolution feature images using the stacked hourglass structure feature extractor and build a multi-level detailed cost volume. We also use the edge of the left image to guide disparity optimization and sub-sample with the low-resolution data, ensuring excellent accuracy and speed at the same time. Furthermore, we design a multi-cross attention model for binocular stereo matching to improve the matching accuracy and achieve end-to-end disparity regression effectively. We evaluate our network on Scene Flow, KITTI2012, and KITTI2015 datasets, and the experimental results show that the speed and accuracy of our method are excellent.

Highlights

The binocular camera plays a significant role in autonomous driving, target detection, and other fields
The purpose of stereo matching is to find the corresponding pixels from the binocular images [5]
In order to test the effectiveness of our multi-level cost volume, we proposed an ablation study to compare the effects of conventional construction methods and our construction methods on network results to prove our design choice

Summary

Introduction

The binocular camera plays a significant role in autonomous driving, target detection, and other fields. It has a series of advantages such as a much lower price than LIDAR, better performance, and fewer errors [1,2]. We can use the binocular camera to achieve excellent depth estimation from a pair of corrected left and right images. The purpose of stereo matching is to find the corresponding pixels from the binocular images [5]. The pixel point (x, y) is in the left image; the same pixel point is (x − d, y) in the right. The depth D of the pixel is fB/d, where f is the focal length of the camera and B is the baseline distance between the center of two cameras [6,7]

Objectives

Methods

Conclusion