多传感器信息预处理约束紧耦合建图算法OA北大核心CSTPCD

In outdoor long-range and large-scale environments, multi-sensor simultaneous localization and mapping ( SLAM) systems often encounter challenges related to insufficient mapping accuracy, drifting, and map crashes. These issues arise due to incorrect sensor fusion, errors in feature matching, and unreliable sensor state information. To mitigate these problems, this paper proposes a factor graph optimization-based algorithm called LVI-SMAD ( tightly-coupled lidar-visual-inertial odometry via smoothing, mapping and DBSCAN ) that addresses tightly-coupled multi-sensor information. It utilizes clustering results obtained from the joint front-end point cloud and visual information as constraints for factor graph optimization. These constraints are incorporated into the output point cloud map of higher frames, enhancing the cohesive relationship between point cloud and visual data. This approach overcomes the inaccurate information matching between lidar and camera. Moreover, the constraints act as supplements when certain sensor information becomes unreliable, thereby reducing positioning drift in scenarios with unstable sensor data and enhancing algorithm consistency. Our experimental results demonstrate the significant performance improvement of LVI-SMAD compared with existing methods. In low-slope and long-span environments, LVI-SMAD reduces absolute trajectory error by 39 . 90％ compared with LVI-SAM and by 63 . 09％ compared with LIO-SAM. Similarly, in high-slope mapping environments, LVI-SMAD achieves a substantial reduction of 41 . 08％ in absolute trajectory error compared to LVI-SAM and a 64 . 87％ reduction compared to LIO-SAM. These results provide compelling evidence for the effectiveness and feasibility of the proposed algorithm.