This paper investigates the use of space-based tracking data to determine the initial orbit of low-earth orbit (LEO) target satellites. A series of tracking data without distance information of a LEO target satellite can be provided by a space-based imaging observer and a target detection process. First, Gauss’ method is applied to a series of a target satellite's tracking data to estimate its state vectors over the observation time span. Due to different input sets and the multiple roots problem of Gauss’ method, a large number of state vector solutions are obtained, and a solution group of the estimated results is proposed. Second, the state vector solutions are grouped according to the multiple roots characteristic of Gauss’ method and then smoothed and fitted based on the motion continuity in each group. Finally, each fitted solution is used to estimate the elements of one orbit, and an optimized result is selected from all the orbit element solutions by mapping and comparing with the original tracking data. Simulation experiments demonstrate the efficacy of our method. Within a short observation time span, the proposed algorithm can obtain an optimized solution for the initial orbit elements of a target satellite.