Semantic Parsing of Street Scene Images Using 3D LiDAR Point Cloud

In this paper we propose a novel street scene semantic parsing framework, which takes advantage of 3D point clouds captured by a high-definition LiDAR laser scanner. Local 3D geometrical features extracted from subsets of point clouds are classified by trained boosted decision trees and then corresponding image segments are labeled with semantic classes e.g. buildings, road, sky etc. In contrast to existing image-based scene parsing approaches, the proposed 3D LiDAR point cloud based approach is robust to varying imaging conditions such as lighting and urban structures. The proposed method is evaluated both quantitatively and qualitatively on three challenging NAVTEQ True databases and robust scene parsing results are reported.