SWITCHED MULTI-MODEL ESTIMATION USING PROBABILISTIC NEURAL NETWORK DECISION FOR MANEUVERING TARGET TRACKING

摘要

Target motion model is usually unknoun dudue to the uncertainty ofmaneuver.or maneuvering target tracking, a typical way is to use the multi-modelM/M)based methods, where multiple filters work in parallel on different motion modelswith diferent state transition functions. However, paralleled working of multiple modelsis unnecessary for target with typical maneuver and remarkably increases the computation complexity and algorithm structure. In this paper, we propose a novel multi-moamethod based on probabilistic neural network (PNN)decision. In the methoatmodel switches among a designed model set and only one model runs during any sampletime. 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 motionmodel 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-MMmethod runs only one filter at any single moment, whuch makes the algorithm structuresimplified and calculation complexity decreased. Simulation results have shoun the improved performance of the PNN-MM method for tracking targets with typical maneuverand less computational Load compared with traditional MM methods.