Toward High-Precision SLAM Ground Truth in GNSS-Denied Large-Scale Environments

This article introduces an efficient and highly accurate framework for generating precise trajectories to evaluate the performance of simultaneous localization and mapping (SLAM) algorithms. Traditional methods for acquiring high-precision SLAM trajectories typically rely on external high-accuracy trajectory references, such as real-time kinematic global positioning systems (RTK-GPSs) or motion capture cameras. However, obtaining high-precision trajectories through conventional means in GPS-denied indoor environments or large underground settings poses a significant challenge. To address this issue, we propose a method of high-precision point cloud matching to calibrate lower-precision point clouds, supplemented by contextual consistency optimization for efficient and accurate trajectory acquisition. We demonstrate the robustness and generalizability of our method using our own dataset, LIVOX-Indoor, with experimental evaluation results showing millimeter-level accuracy.