Research On Multi-model Filtering Algorithm For Tracking Maneuvering Target

Maneuvering target tracking technology is based on radar,sonar and other sensors to observe the target,get some original data information about the maneuvering target,and estimate and predict the moving state of the target by establishing a reasonable motion model and combining various filtering methods.This technology is an important part of modern radar navigation system and plays an extremely important role in military and civil fields.With the rapid development of modern science and technology,people demand higher and higher precision and reliability of navigation system,which puts forward higher requirements for theoretical research and practical application of maneuvering target tracking technology.Based on this,the multi-model filtering algorithm for tracking maneuvering targets is deeply studied in this paper.The specific work is as follows:(1)The interactive multi-model(IMM)algorithm model is established for maneuvering targets with multiple motion modes,which solves the problems of poor tracking accuracy and target loss when tracking maneuvering targets with a single model.At the same time,the influence of model probability transition matrix on IMM tracking performance is explored,and the adaptive transition probability matrix interactive multi-model(ATPM-IMM)algorithm is proposed.The algorithm corrects the model by detecting the change of model probability,and the effectiveness of the algorithm is verified by simulation.(2)Aiming at the defect that IMM algorithm can't adaptively change the model set when tracking the target,this paper puts forward the Adaptive model switching interactive multi-model(AMS-IMM)algorithm for the scene where the target state is not frequently switched and the real-time requirement is high,and the adaptive model collaborative switching interactive multi-model(AMCS-IMM)algorithm for the scene where the target state is frequently switched and the data computation is less.This paper focuses on the way of setting the model sets of maneuvering state and non-maneuvering state by the two algorithms,as well as the transformation rules between the model sets,and gives the design flow of the two algorithms.Finally,the advantages of the two optimization algorithms compared with the fixed model set algorithm are verified by experiments,and their characteristics are summarized.(3)Aiming at the disadvantage of linearization error of extended Kalman filter in IMM algorithm,the GA-BP-IMM algorithm is proposed by fusing the error back propagation(BP)algorithm optimized by GA with IMM algorithm.The working principle of BP neural network algorithm is described in detail,and the defect that it is easy to fall into local minimum is optimized by genetic algorithm,which is applied to the error compensation of interactive multi-models.Finally,the correctness and feasibility of the algorithm are proved by experiments.