A Dynamic Scene Vision SLAM Method Incorporating Object Detection and Object Characterization

Simultaneous localization and mapping (SLAM) based on RGB-D cameras has been widely used for robot navigation in unknown environments. Most current SLAM methods are constrained by static environment assumptions perform poorly real-world dynamic scenarios. To improve the robustness performance of systems environments, this paper proposes a new method indoor scenes object detection. The presented improves ORB-SLAM3 framework. First, we designed an detection module YOLO v5 relied it to tracking accuracy dense point cloud map building was also included, which excludes objects from create with high readability reusability. Full comparison experiments original two representative semantic TUM dataset show that: can run at 30+fps, improved varying degrees compared all four image sequences, absolute trajectory be up 91.10%. is comparable that DS-SLAM, DynaSLAM recent target detection-based algorithms, but runs faster. proposed paper, combines most advanced visual framework, outperforms other terms construction certain reference value navigation, localization, 3D reconstruction.