3-D Point Cloud Map Compression for Connected Intelligent Vehicles

In autonomous vehicles, 3-D point cloud (PCD) maps are widely employed. By matching a point cloud acquired from a 3-D ranging sensor in real time with the PCD map, the ego vehicle can be localized with high accuracy. However, the PCD maps must be compressed and customized to the vehicles because they typically have low computing power, a small memory space, and low-resolution sensors. In this study, we propose an edge service of PCD map compression for connected intelligent vehicles. We overview a general path-aware map compression framework and propose a novel compression method to combine voxelization and a notion of localizability at every waypoint on target paths. Experimental results show that the proposed compression significantly reduces the computational cost at both the edge server and the vehicle while satisfying a required localization performance level.