Detection and classification of pole-like road objects from mobile LiDAR data in motorway environment. (1st December 2017)

• Record Type: Journal Article
• Authors: Yan, Li; Li, Zan; Liu, Hua; Tan, Junxiang; Zhao, Sainan; Chen, Changjun
• Is Part Of: Optics & laser technology. Volume 97(2017)
• Page Start: 272
• Page End: 283
• Publication Date: 2017-12-01
• Subjects: Mobile LiDAR -- Pole-like object -- Detection -- Classification; Optics -- Periodicals; Lasers -- Periodicals; Electronic journals; 621.366
• DOI: 10.1016/j.optlastec.2017.06.015
• Languages: English
• ISSNs: 0030-3992

Abstract

Highlights: A workflow for pole-like road objects detection and classification is presented. The workflow do not require additional information as input except XYZ of point data. The workflow are able to extract pole-like road objects in overlapped region. Abstract: Mobile LiDAR Scanning (MLS) can collect 3-dimensional (3D) road and road-related geospatial information accurately and efficiently. Pole-like objects located in road environment are important street furniture and they are necessary information in road inventory and road mapping. The automatic detection and classification of pole-like road objects from mobile LiDAR data can greatly reduce the cost and improve the efficiency. This paper provides a complete workflow for the detection and classification of pole-like road objects from mobile LiDAR data in motorway environment. The major workflow includes three steps: data preprocessing, pole-like road objects detection and pole-like road objects classification. In data preprocessing step, ground points are removed by an automatic ground filtering algorithm, and then off-ground points are clustered into segments and the overlapped segments containing pole-like road objects are further separated through an iterative min-cut based segmentation approach. In detection step, filters utilizing both prior and shape information are used to detect the target objects. In classification step, features of objects are calculated and classified using Random Forest classifier. Our method was tested on two datasets scanned in motorway environment, and the results showed that the Matthews correlation coefficient of the two datasets in detection step was 93.7% and 95.9% respectively and the overall accuracy of the two datasets in classification step was 96.5% and 97.9% respectively.