视觉SLAM方法综述OA
Survey of Visual SLAM Methods
即时定位与地图构建(SLAM)是机器人感知未知环境的关键技术,是移动机器人、自动驾驶、增强现实等领域的重要组成部分.因此,本文对视觉 SLAM进行系统性综述.首先,简要介绍了视觉 SLAM框架,并通过视觉 SLAM主题数量的变化分析其实际应用的困难及根本原因.然后,介绍了视觉 SLAM常用的数据集和地图类型,并分析了两者具体的应用场景.接着,从传统视觉 SLAM到前沿视觉 SLAM介绍了视觉 SLAM的多种算法及不同类型算法的改进方向及适用场合与局限.进一步地,在传统视觉 SLAM及前沿视觉 SLAM分析的基础上,剖析视觉 SLAM存在的问题以及现行解决方案,将目前存在的问题归类为动态环境的适应性问题、光照变化造成的问题、闭环检测精度不足的问题及计算复杂度与实时性的问题,并在每个问题的最后总结针对上述问题的现行解决方案.最后,对视觉 SLAM方法的未来发展趋势进行展望.
Simultaneous localization and mapping(SLAM)is a key technology for robots to perceive unknown environments and is a vital component in mobile robots,autonomous driving,and augmented reality.Therefore,we systematically review visual SLAM.Firstly,we briefly introduce the visual SLAM framework and analyze the difficulties and root causes of its practical implementation by analyzing trends in the volume of visual SLAM research.Secondly,we introduce the datasets and map types commonly used in visual SLAM,and analyze their specific application scenarios.Thirdly,from traditional visual SLAM to advanced visual SLAM,we introduce various algorithms for visual SLAM,as well as the improvement directions of different types of algorithms and their applicable scenarios and limitations.Further,based on traditional and advanced visual SLAM analysis,we analyze problems in visual SLAM and existing solutions.We classify current problems as adaptability issues in dynamic environments,problems caused by illumination changes,issues of insufficient closed-loop detection precision,and challenges related to computational complexity and real-time performance.At the end of each problem category,we summarize the existing solutions for each respective problem.Finally,the future trend of visual SLAM methods is discussed.
张琳;杨铭;张英杰
西安建筑科技大学信息与控制工程学院,陕西 西安 710055西安建筑科技大学信息与控制工程学院,陕西 西安 710055西安交通大学机械工程学院,陕西 西安 710049
信息技术与安全科学
深度学习特征检测位姿估计闭环检测动态SLAM
deep learningfeature detectionpose estimationclosed-loop detectiondynamic SLAM
《信息与控制》 2026 (4)
625-651,27
陕西省自然科学基金青年基金项目(2025JC-YBQN-865)
评论