Abstract: The constriction of Smart City relies on space foundation which formed by total city awareness and 3D modeling. Furthermore, the automatic recognition and regularization of buildings' outlines are keys for building 3D models. Because the existing regularization methods only use partial information of building outlines, problems such as mistakenly or missing detect the position of inflexions and building outline shapes are not orthogonality after regularization are hard to solve. This article presents a regularization method based on grid filling. This regularization method utilizes the total information of buildings' outline shape, and optimizing the outcome with erosion arithmetic and expansion arithmetic, therefore the regulated orthogonality result can be got without detecting the position of inflexions. Based on the experimental results and analysis, this regularization method can effectively resist noise points and is able to produce the result which has high similarity to buildings' actual outline shape. This method is especially workable for regulating buildings' outline shapes where the outline shapes are extracted from laser radar scanned point cloud and dense matched point cloud by oblique photograph.