基于多源约束自适应视觉SLAM关键帧选取研究

Research on adaptive visual SLAM key frame selection based on multi-source constraints

该文针对现有关键帧选择方法在复杂场景下的稳定性和适应性方面不足问题,提出一种多源约束的自适应视觉SLAM关键帧选取方法。该算法基于相机几何测量原理,设计自适应阈值进行关键帧选取策略;针对复杂环境下的剧烈运动情况,设计基于IMU的实时状态检测机制和熵函数约束标准,进一步提高关键帧选取的稳定性和适应性。在EuRoC数据集和TUM数据集上对该方法进行定性和定量评估。在单目惯性和立体惯性模式下,将估计轨迹与参考轨迹进行对比,以绝对轨迹误差(absolute trajectory error,ATE)、关键帧数量和运行时间作为评判指标,并与ORB-SLAM3方法进行比较。结果显示,提出的方法可显著提高视觉SLAM在复杂环境下的定位精度和稳定性。

This paper introduces an adaptive visual SLAM keyframe selection method with multi-source constraints to enhance the stability and adaptability of existing methods in complex scenes.The algorithm establishes an adaptive threshold for keyframe selection based on camera geometric measurements,incorporates an IMU-based real-time state detection mechanism,and integrates entropy function constraints to further enhance stability and adaptability when dealing with rapid motion in complex environments.The method is evaluated qualitatively and quantitatively using the EuRoC dataset and TUM dataset.Trajectories estimated are compared against reference trajectories in monocular inertial and stereo inertial modes,utilizing metrics such as Absolute Trajectory Error(ATE),number of keyframes,and running time for evaluation alongside a comparison with the ORB-SLAM3 method.Results demonstrate that the proposed method significantly enhances localization accuracy and stability in visual SLAM within complex environments.