Analysis was made to the defects of CS-Jerk model, based on which an improved CS-Jerk algorithm was proposed. According to the statistical characteristics of innovation that the innovation covariance matrix trace is in a chi-square distribution, an activation function was constructed, which could produce activation factor. Then the maximum/minimum acceleration and maneuvering frequency of CS-Jerk model could update adaptively, and thus to adjust the state covariance matrix and filter gain matrix adaptively. Therefore, the error of target state estimation was decreased. Compared with traditional Jerk model and CS-Jerk model, the improved algorithm has higher accuracy in tracking highly maneuvering target. Simulation results show the feasibility of the algorithm.