Enhancing LiDAR–IMU SLAM for Infrastructure Monitoring via Dynamic Coplanarity Constraints and Joint Observation

Real-time acquisition of high-precision 3D spatial information is critical for intelligent maintenance of urban infrastructure. While SLAM technology based on LiDAR–IMU sensor fusion has become a core approach for infrastructure monitoring, its accuracy remains limited by vertical pose estimation drift. To address this challenge, this paper proposes a LiDAR–IMU fusion SLAM algorithm incorporating a dynamic coplanarity constraint and a joint observation model within an improved error-state Kalman filter framework. A threshold-driven ground segmentation method is developed to robustly extract planar features in structured environments, enabling dynamic activation of ground constraints to suppress vertical drift. Extensive experiments on a self-collected long-corridor dataset and the public M2DGR dataset demonstrate that the proposed method significantly improves pose estimation accuracy. In structured environments, the method reduces z-axis endpoint errors by 85.8% compared with Fast-LIO2, achieving an average z-axis RMSE of 0.0104 m. On the M2DGR Hall04 sequence, the algorithm attains a z-axis RMSE of 0.007 m, outperforming four mainstream LiDAR-based SLAM methods. These results validate the proposed approach as an effective solution for high-precision 3D mapping in infrastructure monitoring applications.