This paper proposes a camera-based finger tracking system for no-restrained direct manipulation of virtual objects. The real-time estimation of the occluded finger was realized by using Natural Position (NP) model that is defined as the correlation function among the postures of the neighboring fingers and the Potential Energy (PE) model that describes the anti-mobility of a finger. The virtual object direct manipulation system was developed by combining the tracking system and a general-purpose physical simulation library. The estimation of the occluded fmger positions and the no-restrained direct manipulation of virtual objects were successfully attained.