A tracking algorithm based on improved input estimation is proposed for the complex, variable and non-transient characteristics of maneuvering extended target. Firstly, the Gaussian distribution and inverse Wishart distribution are used to describe the kinematical state and extension of target, respectively, to avoid the complexity of modeling. Secondly, the jerk is introduced into the maneuver detection window to derive the improved input estimation. Then, a Bayesian approach based on improved input estimation is proposed to track the ellipsoidal maneuvering extended target. Finally, simulation results demonstrate that the proposed algorithm can effectively estimate the state and extension information of target, and it has better performance than the conventional input estimation in maneuvering target tracking.