The Research On Neural Network-Based System Of Automatic Incident Detection

With rapid development of traffic, the number of motor vehicles is increasing continuously, and the traffic incidents that happen frequently must be seriously coped with in area of traffic management, so the technique of Automatic Incident Detection(AID) is a hotspot in field of intelligent transportation at present. The goal of Automatic Incident Detection is detecting a incident that has happened on the road in time, thus the incident could be processed in a relatively brief period of time to decrease the loss, and void happening another incident brought by this incident. The core of Automatic Incident Detection is detection algorithm, the algorithm is very important for the whole traffic incidents detection system.In this paper, in the first place, the concept of AID is introduced, the basic principles of representative algorithms are expatiated and the capabilities of all kinds of algorithms are compared. In the second place, the application of Neural Network(NN) in the field of AID is researched, according to Neural Network Forecasting Model of Nonlinear System, the Neural Network Model of Road Traffic is built. Because the BP network has the capability of global approach and there is a model of macroscopical traffic stream, so a kind of approved BP network is used to simulate the model of macroscopical traffic stream, on the other hand a new viewpoint is put forward, that is using difference between real detected values and ouputs of BP network as the criterion to judge whether or not traffic incident has happened. In the final part of this paper, a system of AID is designed by using VC++ and MATLAB, simulation work is also done in this system. By inputting simulation data, capabilities of two kinds of Neural Networks are testified, the result of simulation indicates that this way of combining BP network and RBF network could make traffic incident detecting system have ability of global approach and local approach, and by this way, shortcomings of only using one kind of forwarding network are overcome, basic cause of traffic incidents is reflected more particularly.