融合动态目标跟踪的视觉SLAM算法OA北大核心CSTPCD

To improve the accuracy of simultaneous localization and mapping(SLAM)for robots in dy-namic environments and enable dynamic target tracking,we propose a visual SLAM algorithm that integrates dynamic target tracking.First,we obtain the optical flow vector,instance segmentation results,and depth map information of red-green-blue(RGB)images through a preprocessing mod-ule.Second,we iteratively solve for initial values of camera pose,map points,and dynamic target positions.Finally,we jointly optimize these initial values using an improved factor graph optimiza-tion method.Experiments conducted on the KITTI dataset demonstrate that our algorithm success-fully combines visual SLAM with dynamic target tracking,significantly improving the accuracy of dynamic target tracking and SLAM in dynamic scenes.The overall performance surpasses that of visual dynamic object-aware SLAM(VDO-SLAM).