Railway clearance intrusion detection is critical to the safety of highspeed railway. The foreign object intrusion detection based on 3D laser point cloud segmentation, classification and recognition has the merits of accuracy and intuition, and has broad application prospects in the monitoring of railway key regions such as tunnel entrance and platform. In this paper, an equipment is designed, which drives the 2D laser radar to implement pitching movement and acquires the 3D point cloud of railway scene. Based on the normal consistency principle, the region growing segmentation algorithm is proposed to solve the over segmentation and under segmentation problems caused by Euclidean cluster segmentation and RANSAC segmentation methods. Aiming at the segmented single object point cloud, the Viewpoint Feature Histogram (VFH) is used to extract the 3D point cloud features of different objects; then, based on the VFHs of different objects kdimensional (KD) tree is built, and the closest point searching algorithm is adopted to achieve the classification recognition of single object point cloud. The result of the classification experiment on the typical objects in railway scene shows that the classification recognition accuracy of the proposed algorithm for the typical objects in railway scene is higher than 90%.