A robust target tracking algorithm with fast convergence is very important in single observer passive target tracking system which has inherent disadvantage of weak observability and large initial error. Considering the characteristics of single observer passive target tracking systems via spatial-frequency domain information, a robust improved UKF algorithm is proposed in this paper. Robustness of this algorithm is enhanced by using radial velocity as a state variable and limiting its initial estimating error in a certain bound. A re-sampling process is added to replace the state variables extending of conventional UKF, thus making less computation amount and easier realization. Compared with usual algorithms used in single observer passive target trackings, such as EKF and UKF, the new algorithm is more robust and has faster convergence velocity, lower tracking error. The correctness and validity of the algorithm are comfirmed by the numerical simulation results.