Abstract: The autonomous landing of drones in unknown environments poses a highly challenging task. To solve this problem, a safe and efficient LiDAR landing system is constructed. The proposed system leverages positioning data obtained from the perception information of LiDAR and the inertial measurement unit (IMU), projecting point clouds into an elevation map coordinate system. Utilizing Bayesian generalized kernel elevation inference and an enhanced dynamic point update algorithm, the system predicts, fills, and dynamically updates sparse elevation maps to generate dense and comprehensive elevation maps. Subsequently, a landability analysis map is generated through analyzing terrain geometric parameters in the elevation map such as slope, roughness, and step height. Then, the safest landing position is identified quickly from the landability analysis map by a GPU (graphics processing unit) acceleration method. In addition, an obstacle avoidance strategy based only on single frame point cloud is proposed to avoid the degradation of LiDAR localization near the landing point, and finally achieving the safe landing of the drones. The proposed method achieves excellent autonomous and safe landing results in tests in multiple complex simulation environments and real scenarios.