Autonomous navigation is one of the main enabling technologies for future space missions. While conventional spacecraft are navigated through ground stations, their employment for deep-space CubeSats yields costs comparable to those of the platform. This paper introduces an extended Kalman filter formulation for spacecraft navigation exploiting the line-of-sight observations of visible Solar System objects to infer the spacecraft state. The line-of-sight error budget builds upon typical performances of deep-space CubeSats and includes uncertainties deriving from the platform attitude, the image processing, and the performances of the sensors. The errors due to the low-thrust propagation and light-time delays to the navigation beacons are also taken into account. Preliminary results show the feasibility of the deep-space autonomous navigation exploiting the line-of-sight directions to visible beacons with a 3-sigma accuracy of 1000km for the position components and 2 m/s for the velocity components.