Study On Track Prediction On Multi-sensor Information Fusion

For Multi-sensor information fusion technology for track fusion and prediction,the essence of it is to obtain a more accurate target track with the methods of analysis and synthesis the information of a number of sensors,and establish the corresponding model based on the existing information,to get the future prediction track of the target.Whether the military or civilian,obtain the reliable and accurate track prediction has always been a complex pursuit under the complicated modern society background.This paper,it mainly focuses on three key contents to analyze,and put forward the corresponding solutions which are filter preprocessing,local track prediction and how does the fusion center fuse the local trajectories of each sensor.Firstly,in the actual track fusion system,there are certain measurement errors and noise interference in the track data measured by the sensor.In order to reduce the measurement errors and noise influence on the accuracy of the subsequent prediction and fusion,some filtering preprocessing on the original track data is necessary.For the problem of filtering preprocessing,this paper first introduces the structure of the measured data and the selected coordinate system,due to the fact that the measurement variance of the sensor is not clear for most of time,an asymmetric weighted filtering algorithm is proposed based on the distance-based weighted de-noising filtering algorithm.The algorithm combines the merits of Kalman filtering and distance weighting algorithm,and the simulation results show that the new algorithm is superior to the orginal algorithm in reducing the noise.Secondly,this paper makes a corresponding analysis for the structure of the local track forecast and introduces the idea of time series prediction to predict the sequence of the trajectory.The structure and prediction of fuzzy neural network(FNN)algorithm,which are analyzed in detail.The sliding window method is used to limit the dimension of the network input to the track forecast.At the same time,for the problem of FNN sensitivity to initialization parameters,an improved particle swarm algorithm is used to optimize the FNN algorithm(PSO-FNN),which greatly improves the performance of the algorithm.In order to further solve the problem the FNN is not applicable when the amount of data is low,a hybrid prediction algorithm based on the gray prediction and FNN algorithm weighting ifproposed.The simulation results show that the proposed algorithm has better performance than PSO-FNN.Finally,the trajectory fusion structure model is introduced and the related algorithms of local track fusion are analyzed.Firstly,the algorithm of trajectory fusion of traditional fuzzy neural network is analyzed.In this paper,an improved weighted track fusion algorithm is proposed,which combines the idea of multi-sensor support and can dynamically fuses the local predictive tracks of multiple sensors to produce new target trajectories,and no need to solve the sensor variance compared with the traditional weighted fusion algorithm.By comparing with the traditional FNN fusion algorithm and equal weight weighted fusion algorithm,it is verified that the algorithm has higher fusion precision.