An optimized pose estimation method based on multi-frame data fusion

Traditional single-frame pose estimation methods in simultaneous localization and mapping (SLAM) often suffer from cumulative errors, map misalignment, and trajectory drift due to unreliable inertial measurement unit (IMU) data and sparse point cloud features. To address these issues, this study proposes an enhanced pose estimation method based on multi-frame data fusion and optimized front-end scan matching. The proposed approach performs multi-frame fusion of LiDAR data by exploiting pose transformation relationships between consecutive frames. A weighted LiDAR-IMU fusion strategy is employed for pose prediction. During scan matching, statistical filtering is introduced to remove point cloud noise, and pose estimation is further refined through a secondary matching process. Experimental results demonstrate that, compared to mainstream conventional algorithms, the proposed method improves localization accuracy in real-world scenarios by 28.4%, 30.1%, and 65.3%, respectively, effectively reducing cumulative errors and enhancing trajectory estimation accuracy and mapping quality. This study provides a novel solution for enhancing pose estimation accuracy and mitigating cumulative errors in mobile robots mapping and self-localization tasks.