Freeway Incident Detection Algorithm

Traffic Automatic Incident Detection System(TAIDS),which is one of the key parts of ITS,is able to detect and deal with the traffic incidents on the roads.Many research works are being done in automatic detection for its high detection rate,but low cost and without weather restriction.The incident detection algorithm is the key in automatic.According to the characters of traffic stream,the theories and characteristics of some popular incident detection algorithms are analyzed in this paper.Based on the analysis of their advantages and disadvantages,some algorithms used in the freeway incident detection are presented.The main contribution can be stated as follows:The problem of freeway incident detection is studied by using Radial Basis Function(RBF) Network.The change rule of the traffic flow parameters while happening of incident is analyzed.The dynamic neural network architecture is constructed by selecting the proper traffic flow parameters as inputs.A strategy of merging and deleting the redundant hidden nodes is introduced without prejudice generalization ability of the neural network.The adjustment of hidden nodes is implemented by applying the adaptive learning method,and can make the neural network automatically select learning rate in different training stages.Simulation experiments show that this incident detection algorithm has such advantages as high detection rate and fast learning ability.It is found to be potentially applicable in practice.Support Vector Data Description(SVDD)is used to study the problem of freeway incident.This method optimizes the incident samples and free-flow samples separately,but it only optimizes the dates which have influence to hyper-sphere one time.So it not only holds the advantages of SVDD, and but also gets obtained high training speed.