摘要
随着人工智能技术的发展,深度信息在机器人技术、无人驾驶和虚拟现实等领域的应用日益广泛。然而,现有深度传感器存在数据稀疏和受干扰等问题,生成的深度图限制了其在实际任务中的应用。于是,近年来提出由图像引导的深度补全任务,从稀疏深度图和高质量彩色图像生成密集深度图。在这项任务中,如何融合图像的语义信息和稀疏深度对性能起着重要作用。本文提出了一种将语义信息引入双分支网络中的深度补全框架,以促进语义信息与深度信息的融合,从而提高补全效果。试验结果表明,在加入了语义信息后,本文框架在道路场景数据集上表现优异,验证了其合理性和性能优势。
With the advancement of artificial intelligence technology,the application of depth information in fields such as robotics,autonomous driving,and virtual reality has become increasingly widespread.However,existing depth sensors suffer from issues such as data sparsity and interference,limiting the applicability of the generated depth maps in practical tasks.Consequently,the image-guided depth completion task has been proposed in recent years,aiming to generate dense depth maps from sparse depth maps and high-quality color images.In this task,the integration of semantic information from images and sparse depth data plays a crucial role in performance.This paper proposes a novel depth completion framework that incorporates semantic information into a dual-branch network to facilitate the fusion of semantic and depth information,thereby enhancing the completion results.Experimental results demonstrate that the proposed framework,with the inclusion of semantic information,performs excellently on road scene datasets,validating its rationality and performance advantages.
作者
耿信财
施祥鹏
黄书发
屈文杰
严璐
李必军
杜韵琦
GENG Xincai;SHI Xiangpeng;HUANG Shufa;QU Wenjie;YAN Lu;LI Bijun;DU Yunqi(China Railway Seventh Group Wuhan Engineering Co.,Ltd.,Wuhan 430200,China;State Key Laboratory of Information Engineering in Surveying,Mapping and Remote Sensing,Wuhan University,Wuhan 430079,China)
出处
《测绘通报》
CSCD
北大核心
2024年第S02期191-196,共6页
Bulletin of Surveying and Mapping
基金
国家重点研发项目(2023YFB3907100)
