Robust and Accurate Multi-UAV Cooperative Semantic SLAM Leveraging Lightweight YOLO

Abstract

Multi-UAV visual cooperative SLAM primarily uses cameras as sensors that work together to complete positioning and map generation. However, traditional SLAM systems are based on the assumption of a static environment and face problems of high computational load and poor robustness in dynamic environments. To address these issues, we propose a multi-UAV cooperative semantic SLAM system based on a lightweight You Only Look Once (YOLO) algorithm. First, we made lightweight improvements to the YOLO network with the aim of reducing the system's computational resource consumption and generating semantic information in real time. Second, we proposed an improved feature point screening algorithm that uses object detection and geometric constraints. This algorithm integrates semantic information and geometric constraints to judge the motion state of the target, excluding the feature points of moving targets, thereby improving the robustness and positioning accuracy of the system and making it adaptable to dynamic environments. Finally, we tested and evaluated our approach on the KITTI, TUM, and EuRoc datasets. The experimental results show that our system outperforms current mainstream SLAM systems in terms of feature extraction and positioning accuracy, and has good robustness and real-time performance.