基于三角网格动静区域分离的改进RGB-D SLAM方法（英文）

Improving RGB-D SLAM through extracting static region using delaunay triangle mesh

针对机器人导航中动态环境下的同时定位和建图问题,提出了一种可在动态场景下稳定实时运行的RGB-D SLAM方法。通过从静态背景中分割并剔除动态对象,并在系统RANSAC解算过程中提取静态背景中的特征点来估计相机轨迹,使得系统在动态场景下能稳健定位;对当前帧构建Delaunay三角网格,并判断当前帧与参考帧的匹配点对的距离一致性,通过删除动静状态不一致的点对线段来剔除网格中的动态物体;结合带加权的词典方法,通过减小动态物体在动态场景的权重,进一步提高系统精度。实验结果表明,所提出的方法在TUM数据集的高动态序列中准确度相比现有实时SLAM方法提高了81.37%,显著提高了移动机器人在动态场景下的定位精度。

To solve the problem of simultaneous localization and mapping(SLAM)in the dynamic environment of robotics navigation,a real-time RGB-D SLAM approach that can robustly handle high-dynamic environments is proposed.A novel static region extraction method is used to segment the dynamic objects from the static background,and the feature points in the static region are integrated into the RANSAC method to estimate the camera trajectory.The dynamic entities are identified and isolated by discarding the edges in the Delaunay triangle mesh of current frame according to distance-consistency principle in a rigid body.Combined with the weighted Bag-of-Words method,the system accuracy is further improved by reducing the weight of the dynamic object in the dynamic scene.Experimental results demonstrate that,compared with the existing real-time SLAM method,the proposed method improves the accuracy by 81.37%in the high-dynamic sequences of the TUM RGB-D datasets,which significantly improve the accuracy of navigation and positioning of mobile robots in dynamic scenes.