Traditional data association-based simultaneous localization and mapping （SLAM） methods are prone to causing mismatches between observations and targets， leading to a decrease in pose estimation accuracy. This paper proposes a 3D LiDAR SLAM method for cluttered environments by combining column feature extraction method and random finite set theory based on sequential Monte Carlo implementation. Firstly， stable column features are extracted from segmented point clouds using the M-estimator sample consensus algorithm to obtain static surviving features and new features within a single frame of point cloud data. Subsequently， two types of features are introduced into the RB-PHD-SLAM （Rao-Blackwellized-probability hypothesis density-simultaneous localization and mapping） framework， and the sequential Monte Carlo method is employed to achieve inter-frame propagation of the vehicle’s trajectory probability density and the map posterior. This enables simultaneous estimation of environmental features and vehicle poses. Evaluation results based on both simulation dataset and KITTI dataset show that， compared with the classical FastSLAM algorithm， the proposed method improves the vehicle positioning accuracy by 44.99%， and reduces the average estimation error of feature location and feature number by 49.24% and 56.22%， respectively. These results indicate that the proposed method significantly improves the accuracy and robustness of SLAM， and helps to ensure safe operation of intelligent vehicles.