一种集成语义分割的自监督单目深度估计算法  

A Self-supervised Monocular Depth Estimation Algorithm with Integrated Semantic Segmentation

作　　者：樊寒露 李云栋[1] 

机构地区：[1]北方工业大学信息学院,北京100144

出　　处：《工业控制计算机》2025年第4期87-89,共3页Industrial Control Computer

摘　　要：单幅图像的自监督深度估计是计算机视觉领域近年来热门的研究课题,然而基于重投影损失的纯视觉方法容易受到光照强度变化和运动物体的影响。由于真实世界图像的复杂性,大规模带注释的数据集的获取在时间上和金钱上都非常昂贵,不同数据集间的泛化性也容易受到场景迁移的影响,学习自监督单目深度估计变得更加棘手。因此,提出一种集成语义分割的自监督单目深度估计算法,该算法使用ResNet-50和ImageNet预训练的权重作为主干,并通过一个额外的语义分割分支来进行特征增强,为深度估计提供丰富的上下文信息。Self-supervised depth estimation of a single image has been a hot research topic in the field of computer vision in recent years,but the pure vision method based on reprojection loss is susceptible to the influence of light intensity changes and moving objects,and it is more tricky to learn self-supervised monocular depth estimation.Therefore,a self-supervised monocular depth estimation algorithm with integrated semantic segmentation is proposed,which uses the weights pre-trained by ResNet-50 and ImageNet as the backbone,and performs feature enhancement through an additional semantic segmentation branch to provide rich context information for depth estimation.

关 键 词：深度学习 自监督 语义分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]