Abstract: In the existing methods of indoor 3D model reconstruction, indoor navigation elements that work as space separators are usually regarded as undividable structure. However, the shape difference between two wall surfaces on one wall will cause details loss in the indoor 3D reconstruction room extraction, as well as difficulties in extracting doors and windows. Aiming to solve this problem, this paper proposes an idea of refining space separator. By refining one wall into two wall surfaces, regional growth algorithm is applied to obtain the corner points of inner wall, so that the refined expression of the interior can be obtained. Point cloud densities of corresponding areas on two wall surfaces are compared to avoid the influence of the obstacles blocking the wall surface on the extraction result of door and window extraction. The results show that the proposed method can effectively extract indoor doors and windows, which provides an important basis for the generation of navigation network.