Research into Creation of 3D Point Cloud Map in Large-scale Environments: Improvement of Efficiency in Point Cloud Processing by Grid Management

3D SLAM (Simultaneous Localization and Mapping) is a technique for creating circumstance maps which are usable for measuring the environments and providing navigation information. One of the problems in 3D SLAM is computational efficiency of inserting new points into the circumstance map, since the efficiency is essential for keeping the quality and quantity of the obtained circumstance map and keeping the stability of the mapping procedure. This study focuses on how to improve the computational efficiency of 3D SLAM as a UGV (unmanned ground vehicle) application. We employ a sophisticated and well-organized 3D SLAM package, ETHZASL-ICP-Mapper, as a fundamental implementation. After analyzing the reason why the efficiency becomes decreased through the mapping procedure, we point out main two problems. Then we propose two approaches for overcoming these problems. The first approach is to divide the circumstance maps into sub-grid maps. The second approach is to alternate the k-d tree data structure in the point density control with a voxel grid data structure. The improvements are discussed based on the comparative experimental results.