Visial SLAM algorithm based on YOLOV8 in dynamic scenes

In order to solve the problem that the positioning accuracy of visual SLAM(Simultaneous Localization and Mapping) system is degraded due to the interference of dynamic objects in dynamic environment, this paper proposes a visual SLAM algorithm LYO-SLAM(Light-Weight YOLOv8-seg Optimized SLAM) based on multi feature fusion and dynamic semantic segmentation. Based on ORB-SLAM3, the algorithm integrates LSD(Line Segment Detector) segment detection and ORB(Oriented FAST and Rotated BRIEF) feature point extraction, and by introducing the YOLOv8-seg instance segmentation network, realizes pixel level dynamic region segmentation, generates dynamic masks, effectively eliminates the feature information affected by dynamic objects, and reduces the impact of dynamic interference on back-end optimization, so as to improve the positioning robustness and system stability. The semantic segmentation algorithm based on ORB-SLAM3 proposed in this paper shows good tracking effect on the dynamic data set of TUM RGB-D. Compared with other algorithms, the algorithm studied in this paper has higher accuracy and success rate in dynamic target elimination.