Ground Truth

Small-scale Data Sequence

The small-scale data sequences are recorded inside a 5m x 5m x 3m MoCap arena, which has sub-millimeter accuracy, a frequency of 120Hz, and six DoF. Note that the ground truth presented on the download page are raw MoCap readings, hence further applying the hand-eye calibration result is needed if you want to transform onto the regular camera frame.

Large-scale Data Sequence

Large-scale data sequences are recorded by exerting longer trajectories through larger-scale indoor architectures. We use the Iterative Closest Point (ICP) method to match the motion-compensated LiDAR scans with a pre-scanned dense point cloud of the environment captured by the survey-grade FARO laser scanner, thus resulting in a six DoF ground truth trajectory for the LiDAR frame. We then use CloudCompare, a robust and efficient open-source software to register the LiDAR scans with the high-resolution prior global map. The procedure merely requires an initial guess for the first scan in each sequence, which we set manually. The initial guess of each subsequent scan is then simply set to the optimized pose of the previous scan, thus leading to the complete ground truth trajectory.