Classification of airborne LiDAR point cloud with different feature combinations

Abstract

Automatic classification of point clouds in urban scenes has great application requirements. Different feature dimensions and feature combinations have different effects on the result of point cloud classification. This paper summarizes a series of point cloud feature description methods from multiple perspectives, extracts 22-dimensional point cloud feature vectors, then constructs different combinations of geometric features, point color features and neighborhood color features, and discusses the classification effect of different feature combinations. In order to verify the effectiveness of the feature combination methods, Random Forest (RF) and Extreme Gradient Boosting (XGBoost) machine learning classification models are used for experimental verification and comparative analysis. The results show that the two classification models have good robustness. When only geometric features are used for classification, the F1 score of the two methods is only about 52%, while the overall classification precision of the two methods is improved by more than 20% after combining color features.