Multi-Resources Traffic Information Based Data Fusion Research And Application

The issue of vehicle detection is the basic and key problem in Intelligent Traffic System. Acquiring accurate traffic parameters such as traffic volume and velocity by vehicle detection is the foundation of traffic management such as route guidance and traffic control. Traditional vehicle detection methods usually use only one type of sensor. However, constrained by the characteristic of a certain sensor, there are always some disadvantages for single-type-sensor method to do vehicle detection. The data fusion system with multi-resource traffic information could provide higher accuracy of statistic of velocity, traffic volume and road occupancy, than the single sensor. It will help to realize more efficient traffic management. At present, video sensor has been widely used for vehicle detection, but video sensor is sensitive to the change of light while magnetic sensor is not, so magnetic sensor can be used to complement video sensor in vehicle detection when video is disabled in day-night transformation and complex weather situation to promote the detection accuracy.The research and application of data fusion is based on the former video detection system and magnetic detection system. After study and comparison among different methods of data fusion, an innovative improved data resources reliability based Dempster-Shafer data fusion algorithm is proposed, based on the classical Dempster-Shafer evidence theory. This method not only has the same virtues of low accuracy requirement of probability assignment, richer representation of knowledge, and more similar reasoning logic with human, and also it makes up for the weakness of frequent evidence conflicting problem of classical evidence theory. The experimental results of data fusion, based on the real-time video data and simulated synchronized magnetic data show that the detection results of data fusion are more accurate than that of the single sensor both in day time and night time, especially at night time.A framework of data fusion and fusion center structure are designed based on the former data fusion server development platform. The data fusion framework has two phases, which has the virtues of loose couple and extendable feature, which allows not only magnetic and video signal input but also any other kind of sensor signal input. The fusion center is designed to have separate modules, feature extraction interface, data association unit and data pool, and fusion calculation unit, which is very structural and practical.A fusion protocol is designed to define the format of data package and the interface to analyze the packages. The detecting client is responsible for processing signals and send the results to the fusion server according to the protocol, and the fusion server is responsible for analyze the package and put the data to the data pool, without having to know the specific methods of the detecting client. Then the data association unit and fusion calculation unit process the data in the data pool and give the decision-level fusion result.This research work makes innovative improvement of evidence theory fusion algorithm, fusion experiments on the video data and magnetic data are carried out. The fusion framework and fusion center based on the former data fusion server development platform are designed and realized. It is a helpful attempt in multi-source traffic information fusion and a good foundation for further application of multi-source traffic information fusion.