Hierarchical Road Extraction Using LiDAR Data

LiDAR is a recent and progressive technology for collecting data from surface that operates based on the laser length measurements. High planimetry and altimetry accuracy of the obtained LiDAR point-cloud, as well as the ability to record intensity are the reasons to utilize LiDAR data for detecting objects. Extracting roads as both important urban objects and connection channels of a country is vital significantly. In this paper, a hierarchical approach was proposed for extracting the main road network with acceptable precision. The proposed method eliminated non-road objects by using range and intensity data and applying some filters successively. Also, it prevented to produce gap and fracture in the road network. In this regard, firstly, three features were produced by specifying a threshold on the last intensity pulse and utilizing the last range pulses to obtain nDSM, as well as producing slope with normal vectors. The linear convolution of the produced feature layers was computed to obtain an initial road class. Subsequently, it was tried to remove noises from the initial road network and improve detection results according to the road geometrical characteristics. Finally, the skeleton morphological filter and Fourier features were used to smooth roads boundaries and to eliminate byroads. The evaluation results of the road extraction using our proposed approach achieved 80.56% Correctness and 77.82% Completeness. Generally, we tried to use all parameters that are useful for separating roads from other objects in order to extract the main road network with high accuracy and speed by applying them successively.