Advanced Mapping Using Planar Features Segmented from 3D Point Clouds

Simultaneous localization and mapping has two parallel tasks: localization and mapping. A new approach is introduced in this paper to tackle the mapping problem. Ubiquitous planar surfaces (e.g. walls of buildings along the roads) existing in urban environments are employed as features for mapping. The initial planar surfaces are segmented from the raw point cloud data, by means of the modified selective statistical estimator. Then, planes originating from non-static objects (e.g. large vehicles) and redundant planes resulting from the same building surfaces are removed. Lastly, closely located plane segments that share the same plane equations are properly combined. By this means, small plane segments are merged and useful map features are formed. The constructed map, represented by planar features, is compared with real scenes in Google Map. The comparison results show that the planar features match the building profiles very well. The constructed feature map proves to be accurate, and the proposed mapping method saves storage space by using simple planar features instead of raw point clouds themselves.