An automatic parking system provides convenience for drivers by automatically finding free parking spaces and steering automobiles toward them. This paper proposes a vision-based free parking space detection system. The proposed method consists of two stages. First, the automobile rearview is three-dimensionally reconstructed using the optical flow-based method. The metric information is recovered using the scale ratio between the camera heights in the real world and the reconstructed world. Second, the free parking space is detected by estimating the parking space orientation and the reference point in the top-view of the reconstructed 3D structure. Experimental results show that the proposed method successfully reconstructs the 3D structures of the automobile rearview and detects free parking spaces in the different environments.