Ground-based optical tracking systems used in space surveillance provide no range information, and an observed arc is too short compared to the orbital period to provide enough dynamical information, especially for objects in low Earth orbit. Thus, multiple arcs of tracking data should be accumulated and associated in case of long-range orbit prediction. In this paper, a strategy to associate multiple orbit solutions and obtain accurate orbit solution to predict long-term trajectory using an optical space surveillance system is proposed and analyzed. The strategy consists of three steps: unscented batch estimation, chi-square testing, and element fitting. Each step is validated by numerical simulations, and the practical applicability is verified by applying actual tracking data from the Optical Wide-field Patrol Network (OWL-Net). In both the simulations and the practical cases, the associations are well determined, and the post-fit orbit solution provides a predicted trajectory error of 20 km for a week. The proposed strategy is a standalone process requiring no additional observations or orbit database, and it supports successive tracking by optical equipment with a restricted field of view.