Error Analysis-Based Map Compression for Efficient 3-D Lidar Localization

Large-scale 3-D lidar maps are widely used in mobile robot localization because they can provide excellent constraints. However, the enormous number of point clouds imposes constraints on communication, storage, and computation, which brings a massive demand for localization-oriented point cloud map compression. This article proposes an efficient localization-oriented 3-D lidar map compression algorithm. First, we construct a multipose lidar sampling model based on feasible regions so that the compressed map includes observation data on multiple trajectories. Then, a localization error sensitivity analysis is introduced to score the map points, and their localization contribution is calculated according to the 6-DOF scores and observability of the map points. Finally, according to the localization contribution of map points, multiresolution map compression units and a specific line-to-plane ratio are used to compress the map. We have conducted multiple sets of comparative experiments with our self-recorded multitrajectory dataset to demonstrate the effectiveness and efficiency of our algorithm. Compared with different map compression algorithms, the final results show that when the compression ratio drops to 0.1%, although other algorithms fail, our algorithm can still provide high localization accuracy, which reaches map compression for efficient localization.

Automated summary

This article proposes an efficient localization-oriented 3-D lidar map compression algorithm. The algorithm utilizes a multipose lidar sampling model based on feasible regions to include observation data on multiple trajectories in the compressed map. It also introduces a localization error sensitivity analysis to score the map points and calculates their localization contribution. The algorithm achieves high localization accuracy even when the compression ratio drops to 0.1%.