3D scene reconstruction from 2D images can serve as a foundation for solving a number of robotics and computer vision problems such as robot navigation and 3D object recognition in which covering a large field of view could be critical. As conventional cameras have limited fields of view, we propose an approach to reconstruct large 3D scene models using a combination of a pan-tilt-zoom (PTZ) camera and a moving spherical mirror. The proposed system provides uncertainty estimates of the scene model. The uncertainty of moving spherical mirror detection and localization is represented using samples. Two images are used to computed a location measurement of a 3D point in the scene in which the combinations of the sample sets of the two mirror locations are exploited and location measurements of 3D points are represented using Gaussian distributions. Localization of 3D points in the scene is done using a Kalman filter. The experimental results demonstrate the feasibility of the proposed system.