视觉SLAM运动分割技术综述OA
A Review of Motion Segmentation Techniques for Visual SLAM
作为移动机器人与自动驾驶领域的关键基础技术,视觉同时定位与地图构建(V-SLAM)在动态环境中面临严峻挑战.由动态物体引起的特征匹配错误常常导致定位偏差、地图失真以及系统鲁棒性受损.运动分割技术是提高 V-SLAM性能的重要手段,但在复杂动态场景中准确区分静态和动态元素仍极具挑战性.本文系统梳理 V-SLAM 运动分割研究进展,根据对环境的潜在假设,将现有方法分为三个主要研究范式,并给出各范式的技术原理、代表性策略的核心优势、本质局限及适用边界.最后展望未来的研究方向.
As a critical foundational technology in the fields of mobile robots and autonomous driving,visual simul-taneous localization and mapping(V-SLAM)faces severe challenges in dynamic environments.Feature mismatches induced by dynamic objects frequently lead to localization drift,map distortion,and degradation of system robust-ness.Motion segmentation technology is an important means of enhancing V-SLAM performance,but accurate dis-crimination between static and dynamic elements in complex dynamic scenarios remains highly challenging.This paper systematically reviews the research progress on motion segmentation for V-SLAM.Taxonomically categoriz-ing existing methods into three primary research paradigms based on underlying environmental assumptions,we present the technical principles of each paradigm,along with the core strengths,inherent limitations,and applicabil-ity boundaries of representative strategies.Finally,future research directions are prospected.
冯嘉琪;杨恺伦;林家丞;杨观赐
贵州大学现代制造技术教育部重点实验室 贵阳 550025湖南大学人工智能与机器人学院 长沙 410012贵州大学现代制造技术教育部重点实验室 贵阳 550025||贵州大学贵州省电子信息与智能应用国际科技合作基地 贵阳 550025贵州大学现代制造技术教育部重点实验室 贵阳 550025||贵州大学贵州省电子信息与智能应用国际科技合作基地 贵阳 550025
视觉SLAM动态环境运动分割运动理解多传感器融合移动机器人
visual SLAMdynamic environmentsmotion segmentationmotion understandingmulti-sensor fusionmobile robots
《自动化学报》 2026 (4)
666-692,27
国家自然科学基金(62373116,62473139),贵州省科技计划项目(黔科合支撑[2023]一般 118),湖南省重点研发计划(2025QK3019)资助 Supported by National Natural Science Foundation of China(62373116,62473139),Guizhou Provincial Science and Techno-logy Project(QKHZC[2023]118),and Key R&D Program of Hunan Province(2025QK3019)
评论