Research on a road marking line extraction and automatic classification algorithm
Shiting Yang, Dongliang Zhang, Yening Lu, Jian Wang, Changsheng Li, Yongzhan Yang, Chenglong Guo · 2024
Aiming at the problem of low classification accuracy caused by not considering the local characteristics of point cloud in the current classification of road marks, we proposed a point cloud classification framework based on the eight-neighborhood residual search network model. Firstly, we extract outlet surface cloud based on Cloth Simulation Filter (CSF）and maximum connected region, and use pavement multi-feature image and Ostu segmentation algorithm to extract road marks.Then, we used the method to classify the extraction results for a road marking line point cloud classification algorithm based on eight-neighborhood search residual network. The algorithm acquired local features between adjacent point clouds by adding two local feature sub-extraction blocks, which could improve the final classification accuracy. The experimental results showed that the extraction accuracy of road marker lines reached more than 96%. The classification algorithm proposed in this paper can basically accurately classify five types of road marking lines, including straight right turn, straight left turn, dotted line, solid line and diversion line. Compared with the classic PointNet classification algorithm, the classification accuracy of the proposed algorithm is improved by 6.80% on average.