Recently, computer vision technology has been widely adopted for the safety technology of automobiles. In conventional methods, research on the recognition of obstacles of similar height has advanced, such as the recognition of pedestrians or nearby vehicles. However, we must also help drivers recognize the height of obstacles that the vehicles cannot drive over, such as parking lot sprags and curbstones. We propose a 3D reconstruction method of small obstacles in this paper. Due to the cost and the problem of camera calibration, we use an in-vehicle monocular camera. Generally, a camera-view image is used for 3D reconstruction. For the camera-view image, objects far from the camera are reflected as small and their reconstructed 3D information has less accuracy due to the mismatch that occurs in the template matching process. Therefore, we propose a correct matching method using the shape of the tops of the same object because it does not change the sequential top-view images. Using our proposed method, more correct 3D information can be acquired than with camera-view image.