Effectiveness of Different Machine Learning Algorithms in Road Extraction from UAV-Based Point Cloud

This study presents the evaluation of seven different machine learning (ML) models to classify road surface from point cloud. The study begins with converting two-dimensional images collected from unmanned aerial vehicles (UAV) flights to three-dimensional (3D) point cloud. Seven different ML models, namely, Generalized Linear Model, Linear Discriminant Analysis, Robust Linear Discriminant Analysis, Random Forest, Support Vector Machine with Linear Kemel, Linear eXtreme Gradient Bossting, and eXtreme Gradient Boosting, were developed under different training samples. Finally, road surface were classified from 3D point cloud using developed ML models. To assess the performance of the ML models, manually extracted road surfaces were compared with the ones obtained from ML models. Generalized Linear Model produces the most accurate classification results in a shorter processing time. On the other hand, Linear eXtreme Gradient Boosting and eXtreme Gradient Boosting models produce less accurate road classification in a longer processing time. The classification accuracies of other ML models are between these.