This paper proposed an automatic parking method for the robust, real-time detection and recognition of parking space. By using four on-board fisheye cameras around the vehicle, the bird’s eye view vision system can provide the omnidirectional information of the environment for parking. First, the polynomial fisheye distortion model based method was used to calibrate the cameras, and the image mosaicing method based on the LevenbergMarquardt Algorithm was adopted to combine the four individual images into one. Secondly, the Radon transform based method was employed to extract features of the parking space in the radon space. Finally, the double circular trajectory planning and the preview control strategy were proposed to park the vehicle autonomously. Experimental results show that the proposed method works effectively and robustly in real-time for both automatic parking and parking space recognition.