Efficient Vehicle Segmentation and Classification in Urban Point Clouds Using Modified Euclidean Clustering

The aim of this research is to design and validate methods for detecting specific objects, namely vehicles, in LiDAR point clouds obtained from large‐scale urban mapping. Accurate detection of vehicles in such data is challenging because outdoor environments contain a mixture of objects with varying geometry, density, and reflectivity. The proposed approach focuses on reliable vehicle segmentation and classification while maintaining low computational complexity and high interpretability. The processing pipeline combines terrain filtering, modified Euclidean clustering that incorporates LiDAR intensity information, and PointNet‐based classification. To evaluate the proposed approach under real‐world conditions, a dedicated urban LiDAR dataset (Danubia) containing several hundred vehicles was created using a mobile mapping platform. Experimental results demonstrate that the proposed method achieves a vehicle classification success rate of 92.50%.