Understanding urban environments through 3D point cloud segmentation is critical for applications like autonomous driving and urban planning. This work presents a lightweight, rule-based system to segment and classify point clouds from the KITTI dataset into roads, buildings, trees, and vehicles. This method combines RANSAC for ground plane detection and DBSCAN for clustering, with dynamic thresholding to adapt geometric classification rules to input data. The system achieves 64% overall efficiency, excelling in planar surfaces (e.g., roads at 95% accuracy) but facing challenges with trees and vehicles due to occlusion and shape variability. Unlike deep learning models, our approach requires no training and runs efficiently on standard hardware, making it suitable for real-time applications. The method shows lower accuracy for trees and vehicles in cluttered scenes, highlighting opportunities to improve robustness for complex geometries.