Abstract: To meet the demands of fully autonomous operation of unmanned rotorcrafts, an autonomous safe landing system for unmanned rotorcrafts to land on rugged terrain is developed. The proposed system can automatically analyze the terrain of the landing area through onboard real-time calculation, search feasible landing points and then implement autonomous landing. Mounted with a low-cost stereo RGB-D camera for depth sensing and streaming, the system implements real-time 3D terrain reconstruction of landing area and obtains low-noise depth image of the terrain by utilizing a truncated signed distance function (TSDF) based method. A real-time fine search policy is also specifically designed in order to seek out a suitable landing location according to the shape of the landing gear. Finally, the safe landing of the unmanned rotorcraft is completed by a cascaded PID (proportional-integral-differential) controller. Based on the DJI Matrice 100 quadcopter, the autonomous and safe landing is achieved not only in a customized simulator for algorithm evaluation, but also in a realistic rugged terrain situation. With further optimization, this work can provide safe and effective solutions for unmanned rotorcrafts in emergency landing, delivery and post-disaster rescue.