Efficient Collaborative Mapping Based on Bird’s-Eye View Intelligent Recognition and 3DGS

Multi-UAV collaborative mapping aims to realize fast and stable large-scale scene map construction through the cooperation of multiple UAVs. Existing methods usually use high-precision satellite inertial systems or 3D point cloud registration algorithms to realize the global map construction by multi-aircraft cooperation, which cannot take into account the cost, universality, quality and efficiency of map construction. To solve this problem, an efficient collaborative multi-UAV map building method based on bird’s eye view intelligent recognition and 3D Gaussian Splatting (3DGS) is proposed. Based on the advanced and efficient 3DGS-based aerial scene modeling of a single UAV, we propose a fast and accurate bird’s-eye view recognition algorithm based on Gaussian point cloud plane hypothesis estimation to obtain 2D bird’s-eye view maps by projection mapping for a single scene model. Then we derive the 3D scene registration parameters between different UAVs based on the 2D and 3D correspondence relationship, combined with the bird’s-eye view high-precision matching algorithm, so that we realize the 3D point cloud registration with high precision, high efficiency, and low cost, and obtain the large-scale scene map under the coordination of multiple UAVs. Extensive experimental results on aerial scene datasets show that the proposed method is better than the current advanced point cloud registration algorithms (DCP, PointNetLK, RPM-Net) in terms of speed, performance, and cost, which provides a new solution for multi-UAV collaborative mapping under satellite-free conditions.