A STUDY ON AUTOMATIC EXTRACTION OF 3D POINT CLOUDS BY MACHINE LEARNING FOR ROAD LINES

Abstract

 In Japan, Dynamic maps are used for automatic driving development. Dynamic maps utilize point clouds obtained by MMS, and it is necessary to realize efficient data processing and labor saving. On the other hand, in the past research, the construction of algorithms for automatic extraction is a future subject. In this research, we have studied algorithms for automatically extracting road lane lines from point cloud data. We used the random forest method for a machine learning. In the optimal combination of feature quantity and learning data, accuracy of correct answer rate was 97.7%. The proposed method has high extraction performance even compared with existing methods.