Research On The Multiple Model Maneuvering Target Tracking Algorithm

Research and development of the target tracking technology has been always widely given attention by people due to its significant position and wide application prospect in military and civil area. Many researchers at home and abroad has done deep research on it and got rich results in recent years. In the target tracking system, research on the multiple model target tracking algorithm is also a important direction in current target tracking research area due to its performance of treating structure and parameters with and/or changing and simplifying complex problems. According to setting characteristics of the model set and with or without interacting, the multiple model algorithm can be classified as the static multiple model algorithm, the interacting multiple model algorithm and the variable structure multiple model. Among which, the interacting multiple model algorithm has got more attention because of using multiple motion models to match different target motion models and existing interacting between all models, it can not only improve tracking precision, but also has higher effect-cost ratio in the target tracking system. In recent years, the research points of all researchers in countries mainly centralize on improvement of the model, filtering algorithm and data fusion technology aspects. In the meantime, with continuous development and fast maturity of the new subjects, such as the wavelet algorithm, fuzzy logic theory, genetic algorithm and the neural network, more and more scholars combine the interacting multiple model algorithm with these new subjects, which largely improves the tracking precision and real-time performance of the interacting multiple model algorithm.The Kalman filtering algorithm and commonly used maneuvering target tracking models in the target tracking are firstly studied in this paper, tracking performances of all models based on the Kalman filtering algorithm are simulated and analyzed, and then the tracking performance of the interacting multiple model algorithm is simulated in the same target motion environment and compared with tracking performance of the single model tracking performance according to the simulation results, advantages of the interacting multiple model algorithm compared with the single model tracking algorithm area analyzed and concluded.The Markov probability transfer matrix with prior knowledge is needed in the probability renewing calculation of the interacting multiple model algorithm, but it is difficult to get the prior knowledge. And inconformity between the prior knowledge and current maneuvering target motion state can result large tracking error, even lead to losses of the tracking. Pointing at this disadvantage in the common interacting multiple model algorithm, the interacting multiple model based on the genetic fuzzy neural network (GAFNN-IMM) is proposed in this paper. A fuzzy neural network optimized by the genetic algorithm is designed in this algorithm, and residuals outputting from all filters are putted into the genetic fuzzy neural network. Instead of the model probability calculation in the interacting multiple model algorithm, the matching probabilities of all models in the maneuvering target state fusion and outputting interacting are got according to the self-learning and self adaptive performance of the genetic fuzzy neural network. Tracking performances of the interacting multiple model based on the genetic fuzzy network are analyzed and compared with performance of the common interacting multiple model by means of simulation, which proves that the interacting multiple model based on the fuzzy neural network are superior than common interacting multiple model in aspect of tracking precision. In the last, different target motion environments are set, according to simulation and analysis the tracking error change of the GAFNN-IMM algorithm in different maneuvering target motion environment is compared, tracking characteristics and suitable target tracking environments are analyzed.