Visual SLAM Algorithm of Integrating Dynamic Target Tracking

To improve the accuracy of simultaneous localization and mapping(SLAM) for robots in dynamic 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 module. 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 optimization method. Experiments conducted on the KITTI dataset demonstrate that our algorithm successfully 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).