基于改进ACV模型的视差估计方法  

Disparity Estimation Method Based on an Improved ACV Model

作　　者：秦伦明 余斌 崔昊杨 边后琴 王悉[2] Qin Lunming;Yu Bin;Cui Haoyang;Bian Houqin;Wang Xi(College of Electronics and Information Engineering,Shanghai University of Electric Power,Shanghai 201306,China;School of Electronic and Information Engineering,Beijing Jiaotong University,Beijing 100044,China)

机构地区：[1]上海电力大学电子与信息工程学院,上海201306 [2]北京交通大学电子信息工程学院,北京100044

出　　处：《激光与光电子学进展》2024年第24期131-140,共10页Laser & Optoelectronics Progress

基　　金：国家自然科学基金面上项目(62073024)。

摘　　要：双目立体匹配作为计算机视觉领域的一个热点问题,在距离感知、遥感、自动驾驶等诸多场景中有着广泛的应用。针对目前方法在深度不连续以及边界区域存在视差预测不准确的问题,提出一种基于改进注意力拼接代价体网络的端到端视差估计方法。首先引入多尺度特征融合网络,将浅层与深层包含丰富信息的多尺度特征图融合,增强图像细节的细粒度表示,缓解图像中深度不连续区域视差预测不准确的问题。接着设计Sobel边缘平滑损失,建立视差图边界与场景边缘轮廓对应的约束关系,缓解图像中目标边界区域视差预测不准确的问题。在Sceneflow数据集上的实验结果表明,所提方法在EPE和D1指标上分别达到0.467和1.51%,在KITTI数据集上,所提方法在3-All和D1-All指标上分别达到1.44%和1.61%。与注意力拼接代价体网络相比,所提方法在EPE和D1指标上分别降低3.51%和5.63%,在3-All与D1-All指标上分别降低2.04%和2.42%,可获得更好的视差估计效果。As a hot topic in computer vision,binocular stereo matching has broad applications in various tasks such as distance perception,remote sensing,and autonomous driving.An end-to-end disparity estimation method based on an improved attention concatenation cost volume network is proposed herein to address the challenges of depth discontinuity and inaccurate disparity prediction in boundary regions observed in current methods.First,a multiscale feature fusion network is introduced to combine multiscale feature maps containing rich information from both shallow and deep layers.This approach enhances the fine-grained representation of image details and mitigates the problem of inaccurate disparity prediction in areas with depth discontinuities.Subsequently,a Sobel edge smoothing loss is designed to establish a constraint between the disparity map boundary and the scene's edge contours,alleviating inaccuracies in disparity prediction at the image's target boundaries.Experimental verification of the proposed method on the Sceneflow dataset reveals that the proposed method achieves 0.467 score in the EPE metric and 1.51%in the D1 metric.On the KITTI dataset,the method achieves 1.44%score in the 3-All metric and 1.61%in the D1-All metric.Compared to the attention concatenation cost volume network,the proposed method shows reduced EPE and D1 scores by 3.51%and 5.63%,respectively,and reduced 3-All and D1-All metrics by 2.04%and 2.42%,respectively,demonstrating superior disparity estimation performance.

关 键 词：双目立体匹配 注意力拼接代价体网络 端到端视差估计 多尺度特征融合网络 Sobel边缘平滑损失 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]