A novel registration method is presented for 3D point clouds, acquired from a multi-view camera, for scene reconstruction. In general, conventional registration methods require a high computational complexity, and are not robust for 3D point clouds with a low precision. To remedy these drawbacks, a projection-based registration is proposed. Firstly, depth images are refined based on temporal property by excluding 3D points with large variations, and spatial property by filling holes referring to neighboring 3D points. Secondly, 3D point clouds are projected to find correspondences and fine registration is conducted through minimizing errors. Finally, final colors are evaluated using colors of correspondences, and a 3D virtual environment is reconstructed by applying the above procedure to several views. The proposed method not only reduces computational complexity by searching for correspondences on all image plane, but also enables an effective registration even for 3D points with a low precision. The generated model can be adopted for interaction with a virtual environment as well as navigation in it.