Information Fusion And Its Application In Transportation Surveillance

With the rapid development of sensor technology, a large number of different sensors have been widely used in intelligent transportation systems, in order to achieve real-time detection of a variety of traffic information. However, the problem how to effectively use the mass of information with integration of heterogeneous sensors for intelligent transportation systems has become a serious problem. Thus, multi-source information fusion, which stems from military applications, has been introduced into intelligent transportation, aiming to achieving a satisfactory solution with this issue. Taking Intelligent Traffic Monitoring System, the one of the most important subsystems of Intelligent Transportation Systems, as an application object, this thesis studies information fusion theory and its applications in three different levels.First, for the fusion of data from multi-source heterogeneous sensors in the traffic monitoring systems of urban road networks, this thesis presents a real-time fusion model of heterogeneous traffic information, the Federated Evidence Fusion Model. By combining the Federated Kalman Filter with Evidence Theory fusion model, this model not only performs real-time processing for the Federated Kalman Filter and well deals with uncertain information of Evidence Theory, but also overcomes the deficiencies of Evidence Theory in case of conflict of evidences.Then, the thesis studies the information fusion algorithm of multi-source sensors for the video-based traffic surveillance system. By comparative analysises of the matching performance of the target appearance features, the algorithm carries out a matching model based on multi-feature fusion. On the basis of this matching model, we further propose the strategy of camera topology estimation and target tracking management, and ultimately realize the persistent tracking of multiple pedestrians in a small surveillance network consisting of video cameras.Finally, the thesis studies the algorithm of object detection with foregrounds based multi-feature fusion, for video-based traffic monitoring. Concretely, we realize the video monitoring algorithms of the intermodal freight trains, including the segementation of the container foregrounds, the detection of the container velocity, the container mosiac splicing, then the train type recognition, as well as the distance measurement between two containers.In this thesis, three corresponding fusion platforms are built to validate these three different algorithms, respectively. In the platforms, a great number of actual traffic monitoring data are employed to verify the proposed algorithms. Experiment results reflect their superior performance to the traditional approaches, and indicate that they have broad application prospects.