The automatic classification of massive 3D point cloud obtained by the vehicle laser scanning system is of great significance for target recognition and reconstruction. Traditional point cloud classification requires manual intervention, while most of the existing automatic classification algorithms suffer from low classification accuracy and high computing costs. In this regard, this paper proposes an automatic classification method of vehicle point cloud based on LightGBM. The method first calculates the surface change three-dimensional feature, density feature, elevation feature, and fast point feature histogram of the point cloud. Second, the point cloud normal vector and its neighbors are calculated. The normal vector angle of the domain point and the angle between the normal vector and the horizontal plane are used as constraint features. All the results are combined to obtain a 48-dimensional feature vector. Finally, the LightGBM is constructed to train the point cloud feature vector to complete the classification and prediction. Experiments show that the algorithm can accurately and efficiently complete the automatic classification of vehicle radar point clouds. Compared with the control algorithm, the total accuracy is increased by 8.1% on average, the Kappa coefficient is increased by 18.9% on average, and the calculation time is reduced by 73.7% on average.