Switched Multi-Model Estimation Using Probabilistic Neural Network Decision For Maneuvering Target Tracking

Zhiguo Shi,Yuankai Li,Zeying Wang

DOI: https://doi.org/10.24507/ijicic.14.04.1481

2018-01-01

Abstract:Target motion model is usually unknown due to the uncertainty of target maneuver. For maneuvering target tracking, a typical way is to use the multi-model (MM) based methods, where multiple filters work in parallel on different motion models with different state transition functions. However, paralleled working of multiple models is unnecessary for target with typical maneuver and remarkably increases the computation complexity and algorithm structure. In this paper, we propose a novel multi-model method based on probabilistic neural network (PNN) decision. In the method, target model switches among a designed model set and only one model runs during any sample time. It contains two main steps. First, a model set of target motion is trained and classified by a detector built on PNN. Second, the detector determines the current motion model and based on it uses a standard Kalman filter (KF) for optimal estimation. Different from the way of paralleled working of the usual MM methods, the proposed PNN-MM method runs only one filter at any single moment, which makes the algorithm structure simplified and calculation complexity decreased. Simulation results have shown the improved performance of the PNN-MM method for tracking targets with typical maneuver and less computational load compared with traditional MM methods.