3D Road Boundary Estimation using 3D LiDAR with Scanline-wise 1D Deep Feature and Particle Filtering

Abstract

Recognizing road shape is one of the fundamental functions for outdoor navigation of mobile robots and vehicles. This function is crucial for safe control and used for autonomous navigation combined with global localization using maps or GNSS. This paper describes a method of estimating the 3D structure of road boundaries using a 3D LiDAR with a combination of scanline-wise 1D feature extraction and temporal filtering by particle filter. In outdoor environments, since the road shape changes not on a horizontal plane but three-dimensionally, we model the road boundary shape with a series of 3D segments and estimate its parameters repeatedly with the feature extraction and particle filter. The proposed method is tested in terms of the feature extraction performance and the applicability of autonomous navigation.