Large Airport Scenes Maneuvering Target Tracking Algorithm

The dissertation is focused on the study of maneuvering target tracking on airport surface. As one of key technologies, the study and improvement of interacting multiple model (IMM) algorithm is a main task of the dissertation.Firstly, from modeling and design of filter,basic maneuvering target tracking technologies are introduced. Four popular maneuvering target tracking models and three classical filters including Kalman filtering (KF), extended Kalman filtering (EKF), and particle filtering (PF) are reviewed. The special application and improvement of KF in maneuvering target tracking field are summarized. The basic theory of PF, its defects and corresponding solutions are formulated in detail. Secondly, maneuvering target tracking technology is applied to target tracking on airport surface. By analyzing the background and moving characteristic of target on airport surface, and combining them with Ground Target Tracking technology, single-target tracker is designed. On airport surface, a number of prior information, including road network, target moving restrictions and so on, may be known and used to enhance filtering robustness. To combining the information with filter, three additional data processing modules, including preprocessing of sensor measurements, the model-set adjustment, optimization of filtering output, are taken into account. Theoretical analysis and simulation results show that these modules can improve the tracking performance. In addition, an IMM algorithm including model-set design and adaptation is designed based on airport surface. It is shown by theoretical analysis and simulation results that the robustness of the proposed method is higher than the traditional IMM.Thirdly, a novel approach is proposed for the estimation of likelihood on interacting multiple-model filter. In this approach, the actual innovation, based on a mismatched model, can be formulated as a sum of the theoretical innovation and the distance between matched and mismatched models whose probability distributions are known. The joint likelihood of innovation sequence can be estimated by convolution of the two known probability density functions. The likelihood of tracking models can be calculated by conditional probability formula. Compared with the conventional likelihood estimation method, the proposed method improves the estimation accuracy of likelihood and robustness of IMM, especially when maneuver occurs.Finally, a new interacting multiple model algorithm based on the convolution particle filter is proposed for non-linear target tracking when the distribution of noise is unknown. The algorithm utilizes convolution particle filter to run multiple models in parallel. The previous state posterior probabilities of all models interact each other. Samples from the interacted probability density are regarded as the current initial particles. The outputs of all parallel filters are weighted as system output. Compared with the interacting multiple model algorithm based on particle filter (IMM-PF), the new algorithm improves effectiveness-cost ratio, and eliminates the correlation between the algorithm and analytical probability distribution of measurement noise. The effectiveness of the proposed algorithm is shown by the theoretical analysis and simulation results.