Multi-source Hegerogeneous Traffic Data Fusion Based On Federal Kalman Filter

With the continuous increase in the number of motor vehicles in our country,the degree of congestion on urban roads has become increasingly serious.At present,the Intelligent Traffic System(ITS)uses a variety of sensors to be laid on the road surface and in the air of urban roads to form a three-dimensional sensor network that can monitor traffic such as traffic flow and speed in real time to provide a road forecast for traffic travelers.Due to the large number of sensors involved in traffic,the variety of types,the different adaptation conditions have a great impact on its measurement accuracy.Therefore,this paper uses a method to fuse the detection data of three types of traffic detectors to improve the accuracy of final result.As a more classical algorithm,the federated Kalman filter algorithm belongs to distributed Kalman filter.It has small computational complexity and high stability.It is often used in various fields of data fusion and can effectively predict data.This article has done the following main work:Firstly,this paper discusses the extensive application of data fusion technology in intelligent transportation system,investigates the current situation of data fusion field at home and abroad,describes the latest research results and the future development.By contacting the real application scenarios,we propose a fusion algorithm that can integrate the three kinds of traffic sensor data based on the feature layer.And we also determine the traffic flow and the average vehicle speed prediction as our research goal.Secondly,five common data fusion methods including Bayesian theory,neural network,cluster analysis,principal component analysis and Kalman filter are analyzed.Their advantages and disadvantages are compared.The Federated Kalman filter algorithm is proposed as data fusion method.Thirdly,we discuss the three kinds of traffic sensors(loop detector,microwave detector,video detector)in detail,including their advantages and disadvantages.In addition,the commonly used traffic data preprocessing methods are analysed,and the real traffic data is processed by data smoothing prediction.Finally,a distributed Kalman filter model based on the traffic flow and vehicle average speed of the above three traffic detectors is designed.The real vehicle traffic and vehicle average speed data of Hangzhou Stadium Road in Zhejiang Province are selected.The simulation shows that the relative prediction error is reduced.The fusion result is more accurate than the detection value of the three sensors compared with the mean fusion method.Thus,the feasibility of the method was verified.