Identification of pole-like objects from mobile laser scanning data of urban roadway scene

The information acquisition about pole-like objects (PLOs) situated along the road is important in roadway inventory related studies, such as road safety analysis and street visibility studies for traffic management. To collect the detailed roadway information, mobile laser scanning (MLS) system has been adopted as a mainstream tool since last decade. In this paper, a machine learning-based approach using random forest is proposed to identify PLOs in MLS data of roadway scene. First, ground points are removed from the input data thereafter the non-ground points are clustered into cluster features. The random forest-based model is trained using dimension and Eigen values-based feature variables derived from the training samples of various PLOs and nonPLOs features manually extracted from the training MLS data. The proposed method was tested on two different MLS datasets, which were acquired along 4.2 and 2 km long urban roadway environment having perfect as well as complex roadway scenes. Both, simple as well as complex PLOs were successfully identified in these datasets and an average correctness and completeness were obtained 97.67% and 97.79%, respectively. Therefore, the proposed approach has potential to deliver promising results in perfect as well as complex roadway surroundings.