Freeway Accident Detection Model Based on Support Vector Machine

Abstract

The key technology of freeway accident detection was studied in order to set up a quick and efficient accident detection system and promote the efficiency of accident rescue. On the basis of characteristic analysis of the existing models, the freeway accident detection model based on support vector machine (SVM) theory was put forward. With database established by self-developed EAD-Simulations system, a simulation experiment was applied to the model. The effects of different kernel functions on detection performance were analyzed and the performance indexes, such as upstream input, upstream and downstream input and different input of features combination were studied. The results show that the excellent performances of the model are demonstrated by contrast with California model. The detection rate raises 179%; error detection rate drops at 0.50% and average detection time cuts down 81%. In addition, the optimal input characteristic combined by occupancy and flow rate in upstream is received.