The invention provides a vision-based self-aid landing navigation system for an unattended air plane, which comprises a software arithmetic and a hardware arrangement. The software arithmetic comprises a computer vision arithmetic and an information merging and state estimation arithmetic; the hardware arrangement comprises a runway feature, on-board sensor subsystem and an information merging subsystem. The on-board sensor subsystem comprises an on-board camera system, an on-board inertial navigation system, a height meter system and a magnetic compass, measures the real UAV state, traces and analyzes the runway features by the on-board camera system, gets the measurements of the runway feature points, and sends the measurements to the information merging subsystem. The information merging subsystem gets the estimated value of the runway feature points through the runway model and the camera system model based on the estimated value of the airplane state at previous period as well as the measurement on the airplane state at current period by the on-board sensor subsystem, and compares with the measurements, merges other measurement information, and gets navigation information of high accuracy through calculation by the data processing module.