Kohei MATSUZAKI
Hiromasa YANAGIHARA
This paper focuses on the problem that the data volume of a 3D point cloud map becomes enormous in scan matching, which is one of the methods for estimating the self-position of an autonomous vehicle. Thus, this paper proposed a map generation method and a self-position estimation method for three-dimensional point cloud maps. The originality of this paper is high. The novelty is that the three-dimensional point cloud is represented by a voxel grid, and the sub-volume of the set is generated as vector data by vector quantization. The data quantity is significantly reduced as compared with the method of individually quantizing the point data. As a result of compressing the map data, the storage capacity of the map data has been reduced, and the map data has been distributed over the same network bandwidth as voice data. Further, in self-position estimation, the similarity between the map data and scan data is calculated, and the estimation accuracy of the self-position is set to the same level as the conventional one. As described above, it is a valuable paper in that both data compression and maintaining estimation accuracy are compatible. In this paper, the focus is on the automatic driving of vehicles, but the compression of map information and self-location estimation can be applied to other fields. These can include patrol service in a building by robots and home delivery service flying in the sky. Thus, the proposed method has the potential to contribute to various industries and is expected to contribute in the real world. This paper has high utility and novelty and can be highly evaluated as a paper suitable for the Society's Best Paper Award.
