Study And Performance Analysis Of Maneuvering Target Tracking Algorithms Based On Model-Set Adaptive IMM Approach

As the improvement of modern technology, the complexity and maneuverability of maneuvering targets intend to be higher, which raises greater requirement to tracking accuracy, robustness and real-time performance of maneuvering targets tracking algorithms. Traditional tracking algorithms that based on a single model can’t achieve the new requirement, so multiple model algorithms based on Kalman filter become an effective access to resolve this problem. However, kinds of multiple algorithms still need to be improved in robustness and the match of model-sets. Thus, it is very important for both theory and application to study maneuvering target tracking algorithms based on model-set adaptive IMM (Interacting Multiple Model) filtering approach.Aiming at solving the underflow problem of model-set selection probabilities in model-set adaptive IMM algorithms, firstly, this paper studies the likelihood function value reduction mechanism in Novel-IMM algorithm when model-sets mismatch with target modes, and addresses what makes the underflow problem happen in model-set selection probabilities. Based on this, an underflow prevented selection probabilities (UPSP) algorithm is proposed. Besides, this paper defines the mathematical definition and flowchart of this algorithm. The effectiveness of UPSP algorithm is demonstrated by changes of tracking accuracy and model-set selection probabilities. Furthermore, this paper proposes the probability threshold scheme in UPSP algorithm. Then, to acquire the mathematical definition of threshold value, an exponent implemented model-set adaptive IMM algorithm (EAIMM) is proposed based on the correlation between model-set likelihood function and model-set selection probabilities. The EAIMM algorithm not only resolves the underflow problem in standard Novel-IMM, but also becomes more sensitive in tracking target mode changes.On the base of UPSP algorithm, this paper presents an fast model-set adaptive IMM (FAIMM) algorithm that employs the steady state Kalman filter to reduce computational burden of CV and CA model based on UPSP approach, which improves the cost-effectiveness of Novel-IMM. The FAIMM algorithm reduces the executing time by40%while keeping similar tracking accuracy compared with UPSP based Novel-IMM. Finally, the FAIMM-EV algorithm is derived by incorporating Extended Viterbi strategy into FAIMM. The FAIMM-EV algorithm could avoid excessive model competition to some extent. Simulation result reveals that the FAIMM-EV algorithm reduces algorithm complexity by35%compared with Novel-IMM algorithm and has higher tracking accuracy than FAIMM algorithm, which resolves the poor cost-effective problem in model-set adaptive IMM algorithm.This paper demonstrates the proposed four algorithms based on two maneuvering target tracking scenes. Simulation results reveal that the proposed algorithms have advantages in tracking accuracy, robustness, and cost-effectiveness. Thus, the proposed algorithms can be considerable alternatives of standard Novel-IMM in real-time application.