This survey reviews feed-forward,point-cloud 3D reconstruction methods from DUSt3R to VGGT and their recent variants.Here,feed-forward primarily refers to predicting dense geometry and,when applicable,camera poses thr...This survey reviews feed-forward,point-cloud 3D reconstruction methods from DUSt3R to VGGT and their recent variants.Here,feed-forward primarily refers to predicting dense geometry and,when applicable,camera poses through learned network inference,without relying on classical per-scene SfM+MVS optimization as the main inference mechanism.We first formalize the reconstruction task in pose-aware and pose-free settings,and contrast feed-forward point-map regression with classical Structure-from-Motion and Multi-View Stereo pipelines.Building on this,we organize existing methods into three stages:early pairwise models typified by DUSt3R,DUSt3R-style extensions that enhance multi-view consistency,streaming,efficiency,and dynamic-scene handling,and large unified transformers such as VGGT that process tens to hundreds of views jointly,while noting differences in their inference paradigms.We analyze these models along shared axes,including scene representation,correspondence reasoning,pose regression,fusion strategies,and the role of large-scale training data.We summarize widely used 3D datasets and evaluation metrics,and provide a case study on the DTU benchmark for multi-view depth and point map estimation,highlighting accuracy-efficiency trade-offs between optimization-based and feed-forward approaches.Finally,we discuss open challenges in data scarcity,sparse-view reconstruction,non-Lambertian structures,dynamic scenes,long-context processing,and resource-efficient deployment,and outline future directions that combine feed-forward architectures with differentiable rendering,generative priors,and safety mechanisms to enable scalable and trustworthy 3D reconstruction systems.展开更多
当今主流地图构建系统由于定位精度不高、重投影误差较大等问题,限制了稠密地图的生成。尤其在动态场景中,系统的实时性和地图的高精度之间无法共存,以及物体的往复移动为后续地图精度的提升带来了额外的困难。针对上述问题,提出了一种...当今主流地图构建系统由于定位精度不高、重投影误差较大等问题,限制了稠密地图的生成。尤其在动态场景中,系统的实时性和地图的高精度之间无法共存,以及物体的往复移动为后续地图精度的提升带来了额外的困难。针对上述问题,提出了一种基于闭环检测和自适应降采样的视觉SLAM点云地图构建方法(Visual SLAM point cloud map construction method based on closed-loop detection and adaptive downsampling,PCL-LCAD)。上述方法从视觉SLAM系统建图的角度出发,加入3D点云技术,构建一个闭环检测优化模型,扩大生成地图的面积,再建立一个点云自适应降采样模型,利用KD-tree算法对其体素滤波进行改进。实验结果表明,PCL-LCAD方法能在保障准确性和实时性的同时,降低地图占用空间并且提高地图稠密度。展开更多
针对弱纹理和变光照环境下基于点特征的视觉SLAM(simultaneous localization and mapping)算法轨迹漂移的问题,提出了一种基于改进自适应阈值ELSED算法(Adaptive-ELSED)的快速点线融合双目视觉SLAM算法。通过在ELSED算法中添加自适应阈...针对弱纹理和变光照环境下基于点特征的视觉SLAM(simultaneous localization and mapping)算法轨迹漂移的问题,提出了一种基于改进自适应阈值ELSED算法(Adaptive-ELSED)的快速点线融合双目视觉SLAM算法。通过在ELSED算法中添加自适应阈值矩阵,动态调整不同光照条件下梯度阈值,并使用长度抑制和短线合并策略,提高线特征的质量。利用基于双目几何约束和图像结构相似性(SSIM)进行快速线段特征三角化。基于历史位姿及误差分析获取初始位姿,通过自适应因子实现光束法平差过程中点线特征的更有效融合。实验结果表明,所提算法在提高线特征质量的同时,耗时仅为LSD算法的50%,线特征匹配速度较传统LBD算法提升67%,挑战性场景下轨迹误差较ORB-SLAM3降低62.2%,系统的平均跟踪帧率为27帧/s,在保证系统实时性的同时,显著提升了系统在弱纹理、变光照环境下的精度和鲁棒性。展开更多
视觉同步定位与建图(simultaneous localization and mapping,SLAM)是实现移动机器人自主定位并构建环境地图的关键环节。SLAM技术虽能精确重建环境几何结构,却难以为机器人提供执行复杂任务所需的语义理解能力;建筑信息模型(building i...视觉同步定位与建图(simultaneous localization and mapping,SLAM)是实现移动机器人自主定位并构建环境地图的关键环节。SLAM技术虽能精确重建环境几何结构,却难以为机器人提供执行复杂任务所需的语义理解能力;建筑信息模型(building information model,BIM)包含丰富的建筑信息,但与机器人操作系统(robot operating system,ROS)之间存在显著的数据格式和表达方式差异,且现有研究多采用人工方式进行转换,效率低下难以规模化应用,且室内环境并非静态不变,从而会影响机器人的导航决策。因此,提出一种集成BIM数据的ROS室内语义地图构建与动态更新方法。通过研发工业基础类(industry foundation classes,IFC)到统一机器人描述格式(unified robot description format,URDF)自动转换器,实现从BIM到机器人仿真环境的自动化建模;融合YOLOv8与随机采样一致性(random sample consensus,RANSAC)算法,建立视觉驱动的语义地图动态更新机制。结果表明,静态建筑元素还原准确率达98%以上,动态物体识别精度达0.9以上,显著提升了语义地图的自动化程度、知识丰富度及环境适应性。展开更多
基金Supported by the National Natural Science Foundation of China(Grant Nos.62595774,62302297,72192821,62272447,62472285,62472282)the Fundamental Research Funds for the Central Universities(Grant Nos.YG 2023 QNB 17,YG 2024 QNA 44)+1 种基金the National Key R&D Program of China(Grant No.2024 YFE 0115500)the Beijing Natural Science Foundation(Grant No.L 222117).
摘要This survey reviews feed-forward,point-cloud 3D reconstruction methods from DUSt3R to VGGT and their recent variants.Here,feed-forward primarily refers to predicting dense geometry and,when applicable,camera poses through learned network inference,without relying on classical per-scene SfM+MVS optimization as the main inference mechanism.We first formalize the reconstruction task in pose-aware and pose-free settings,and contrast feed-forward point-map regression with classical Structure-from-Motion and Multi-View Stereo pipelines.Building on this,we organize existing methods into three stages:early pairwise models typified by DUSt3R,DUSt3R-style extensions that enhance multi-view consistency,streaming,efficiency,and dynamic-scene handling,and large unified transformers such as VGGT that process tens to hundreds of views jointly,while noting differences in their inference paradigms.We analyze these models along shared axes,including scene representation,correspondence reasoning,pose regression,fusion strategies,and the role of large-scale training data.We summarize widely used 3D datasets and evaluation metrics,and provide a case study on the DTU benchmark for multi-view depth and point map estimation,highlighting accuracy-efficiency trade-offs between optimization-based and feed-forward approaches.Finally,we discuss open challenges in data scarcity,sparse-view reconstruction,non-Lambertian structures,dynamic scenes,long-context processing,and resource-efficient deployment,and outline future directions that combine feed-forward architectures with differentiable rendering,generative priors,and safety mechanisms to enable scalable and trustworthy 3D reconstruction systems.
摘要当今主流地图构建系统由于定位精度不高、重投影误差较大等问题,限制了稠密地图的生成。尤其在动态场景中,系统的实时性和地图的高精度之间无法共存,以及物体的往复移动为后续地图精度的提升带来了额外的困难。针对上述问题,提出了一种基于闭环检测和自适应降采样的视觉SLAM点云地图构建方法(Visual SLAM point cloud map construction method based on closed-loop detection and adaptive downsampling,PCL-LCAD)。上述方法从视觉SLAM系统建图的角度出发,加入3D点云技术,构建一个闭环检测优化模型,扩大生成地图的面积,再建立一个点云自适应降采样模型,利用KD-tree算法对其体素滤波进行改进。实验结果表明,PCL-LCAD方法能在保障准确性和实时性的同时,降低地图占用空间并且提高地图稠密度。
摘要针对弱纹理和变光照环境下基于点特征的视觉SLAM(simultaneous localization and mapping)算法轨迹漂移的问题,提出了一种基于改进自适应阈值ELSED算法(Adaptive-ELSED)的快速点线融合双目视觉SLAM算法。通过在ELSED算法中添加自适应阈值矩阵,动态调整不同光照条件下梯度阈值,并使用长度抑制和短线合并策略,提高线特征的质量。利用基于双目几何约束和图像结构相似性(SSIM)进行快速线段特征三角化。基于历史位姿及误差分析获取初始位姿,通过自适应因子实现光束法平差过程中点线特征的更有效融合。实验结果表明,所提算法在提高线特征质量的同时,耗时仅为LSD算法的50%,线特征匹配速度较传统LBD算法提升67%,挑战性场景下轨迹误差较ORB-SLAM3降低62.2%,系统的平均跟踪帧率为27帧/s,在保证系统实时性的同时,显著提升了系统在弱纹理、变光照环境下的精度和鲁棒性。
摘要视觉同步定位与建图(simultaneous localization and mapping,SLAM)是实现移动机器人自主定位并构建环境地图的关键环节。SLAM技术虽能精确重建环境几何结构,却难以为机器人提供执行复杂任务所需的语义理解能力;建筑信息模型(building information model,BIM)包含丰富的建筑信息,但与机器人操作系统(robot operating system,ROS)之间存在显著的数据格式和表达方式差异,且现有研究多采用人工方式进行转换,效率低下难以规模化应用,且室内环境并非静态不变,从而会影响机器人的导航决策。因此,提出一种集成BIM数据的ROS室内语义地图构建与动态更新方法。通过研发工业基础类(industry foundation classes,IFC)到统一机器人描述格式(unified robot description format,URDF)自动转换器,实现从BIM到机器人仿真环境的自动化建模;融合YOLOv8与随机采样一致性(random sample consensus,RANSAC)算法,建立视觉驱动的语义地图动态更新机制。结果表明,静态建筑元素还原准确率达98%以上,动态物体识别精度达0.9以上,显著提升了语义地图的自动化程度、知识丰富度及环境适应性。