Aimed at the problem of "miss" in the traditional GPS navigation system, a new algorithm for selflocalization is presented by taking the visual navigation as a novel navigation manner without compute camera calibration and any remote sensory information, and the position can be well estimated on the property of projective geometry and relation of geometry. Firstly, the target in the plane is located with the target's position to realize the vision selflocalization. The conclusion demonstrates the effectiveness of the model. The experiments show that the method can effectively eliminate the position, and the maximal different value is 0.01 m. A novel thoughtway with simplicity, real time and high precision is provided for vision selflocalization of UAV is provided．