RefDC:聚焦远距离深度特征细化的深度补全方法

doi:10.13474/j.cnki.11-2246.2026.0611

摘要/Abstract

摘要： [目的] 针对双目-激光雷达融合特征中远距离物体深度信息少,以及现有深度补全模型在视差-深度转换过程中的误差累积放大问题,本文提出了一种聚焦远距离深度特征细化的深度补全框架——RefDC。[方法]首先,优化了深度体表示方法,以真实度量深度为体素维度进行体表示构建,并通过软编码策略提高深度精度,以增强深度细节特征;然后,提出了基于迭代假设引导的深度细化模块,以提取全局与局部的语义、几何特征,并通过轻量级ConvGRU迭代细化深度结果,逐步修正累积误差。[结果]在真实与虚拟的数据集上,RefDC均在精度上取得了最先进的水平,其中整体性能分别提升4.2%与11.9%,在远距离深度补全上提升6.0%。[结论]与传统方法相比,RefDC能够更精确地捕捉远距离深度特征,并显著降低了计算负担。

关键词: 深度学习, 深度补全, 远距离深度细化, 双目-激光雷达融合, 多模态融合

Abstract: [Purposes]In response to the limited depth information of distant objects in stereo-LiDAR fusion features and the issue of error accumulation and amplification in the disparity-to-depth conversion process of existing depth completion models,this paper proposes a depth completion method—RefDC,focused on long-range depth feature refinement.[Methods]Firstly,this paper optimizes the depth voxel representation by constructing a volumetric representation using true metric depth as the voxel dimension,and enhance depth precision through a soft encoding strategy,improving depth detail features.Then,this paper introduces an iterative hypothesis-guided depth refinement module to extract global and local semantic and geometric features.Using a lightweight ConvGRU,the module iteratively refines the depth results and gradually corrects accumulated errors.[Findings]On both real and synthetic datasets,RefDC achieves state-of-the-art accuracy,with overall performance improvements of 4.2%and 11.9%,and a 6.0%enhancement in long-range depth completion.[Conclusions]Compared to traditional methods,RefDC can capture long-range depth features more accurately while significantly reducing computational burden.

Key words: deep learning, depth completion, long-range depth refinement, stereo-LiDAR fusion, multi-modal fusion

中图分类号:

P237

杜韵琦, 施泓羽, 张红娟, 李必军. RefDC:聚焦远距离深度特征细化的深度补全方法[J]. 测绘通报, 2026, 0(6): 67-73.

DU Yunqi, SHI Hongyu, ZHANG Hongjuan, LI Bijun. RefDC: depth completion method focused on long-range depth feature refinement[J]. Bulletin of Surveying and Mapping, 2026, 0(6): 67-73.

导出引用管理器 EndNote|Reference Manager|ProCite|BibTeX|RefWorks

链接本文: http://tb.chinasmp.com/CN/10.13474/j.cnki.11-2246.2026.0611

http://tb.chinasmp.com/CN/Y2026/V0/I6/67

参考文献

[1] 郑柏伦,冼楚华,张东九.融合RGB图像特征的多尺度深度图像补全方法[J].计算机辅助设计与图形学学报,2021,33(9):1407-1417.
[2] 陈诚,张永博,李必军.激光点云在无人驾驶路径检测中的应用[J].测绘通报,2016(11):67-71.
[3] 吴宇晖,李晓娟,刘越.面向增强现实的虚实遮挡技术综述[J].中国图象图形学报,2024,29(10):2859-2879.
[4] 王岩,张婷婷,王飞,等.一种基于深度图改进的分块密集匹配方法[J].测绘通报,2023(2):78-83.
[5] ZHAO Zixiang,ZHANG Jiangshe,GU Xiang,et al.Spherical space feature decomposition for guided depth map super-resolution[C]//Proceedings of 2023 IEEE/CVF International Conference on Computer Vision (ICCV).Paris:IEEE,2024:12513-12524.
[6] TANG Jie,TIAN Feipeng,AN Boshi,et al.Bilateral propagation network for depth completion[C]//Proceedings of 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Seattle:IEEE,2024:9763-9772.
[7] LI Ang,HU Anning,XI Wei,et al.Stereo-LiDAR depth estimation with deformable propagation and learned disparity-depth conversion[C]//Proceedings of 2024 IEEE International Conference on Robotics and Automation (ICRA).Yokohama:IEEE,2024:2729-2736.
[8] YANG Guanglei,TANG Hao,DING Mingli,et al.Transformer-based attention networks for continuous pixel-wise prediction[C]//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision (ICCV).Montreal:IEEE,2022:16249-16259.
[9] 张久伶,朱美玲,徐馨润,等.单目度量深度估计综述[J].计算机系统应用,2025,34(8):1-13.
[10] MAI N A M,DUTHON P,KHOUDOUR L,et al.Sparse LiDAR and stereo fusion (SLS-fusion)for depth estimation and 3D object detection[J].IET Conference Proceedings,2021,2021(1):150-156.
[11] CHOE J,JOO K,IMTIAZ T,et al.Volumetric propagation network:stereo-LiDAR fusion for long-range depth estimation[J].IEEE Robotics and Automation Letters,2021,6(3):4672-4679.
[12] LIPSON L,TEED Z,DENG Jia.RAFT-stereo:multilevel recurrent field transforms for stereo matching[C]//Proceedings of 2021 International Conference on 3D Vision (3DV).London:IEEE,2022:218-227.
[13] XU Gangwei,WANG Xianqi,DING Xiaohuan,et al.Iterative geometry encoding volume for stereo matching[C]//Proceedings of 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Vancouver:IEEE,2023:21919-21928.
[14] CHANG Jiaren,CHEN Yongsheng.Pyramid stereo matching network[C]//Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Salt Lake City:IEEE,2018:5410-5418.
[15] GEIGER A,LENZ P,URTASUN R.Are we ready for autonomous driving？ The KITTI vision benchmark suite[C]//Proceedings of 2012 IEEE Conference on Computer Vision and Pattern Recognition.Providence:IEEE,2012:3354-3361.
[16] CABON Y,MURRAY N,HUMENBERGER M.Virtual KITTI 2[EB/OL].(2020-01-29)[2025-02-20].https://arxiv.org/abs/2001.10773.
[17] XU Haofei,ZHANG Jing,CAI Jianfei,et al.Unifying flow,stereo and depth estimation[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2023,45(11):13941-13958.
[18] ZHANG Youmin,GUO Xianda,POGGI M,et al.CompletionFormer:depth completion with convolutions and vision transformers[C]//Proceedings of 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Vancouver:IEEE,2023:18527-18536.
[19] WANG Yufei,LI Bo,ZHANG Ge,et al.LRRU:long-short range recurrent updating networks for depth completion[C]//Proceedings of 2023 IEEE/CVF International Conference on Computer Vision (ICCV).Paris:IEEE,2024:9388-9398.
[20] YAN Zhiqiang,LIN Yuankai,WANG Kun,et al.Tri-perspective view decomposition for geometry-aware depth completion[C]//Proceedings of 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Seattle:IEEE,2024:4874-4884.
[21] WANG Kun,YAN Zhiqiang,FAN Junkai,et al.Learning inverse Laplacian pyramid for progressive depth completion[EB/OL].[2025-02-11].https://arxiv.org/abs/2502.07289.
[22] XU Zhenyu,LI Yuehua,ZHU Shiqiang,et al.Expanding sparse LiDAR depth and guiding stereo matching for robust dense depth estimation[J].IEEE Robotics and Automation Letters,2023,8(3):1479-1486.
[23] ZHANG Yongjian,WANG Longguang,LI Kunhong,et al.SLFNet:a stereo and LiDAR fusion network for depth completion[J].IEEE Robotics and Automation Letters,2022,7(4):10605-10612.
[24] KIM J,SHIN U,HEO S,et al.Exploiting cross-modal cost volume forMulti-sensor depth estimation[C]//Proceedings of 2024 Computer Vision.Singapore:Springer,2025:350-366.