Back-Propagation Neural Network for Traffic Incident Detection Based on Fusion of Loop Detector and Probe Vehicle Data

Abstract

Traffic incident detection based on a fusion of various available data sources has been an evolving research topic in ITS. This paper proposes a data fusion model for traffic incident detection using BP neural network. In this model, the cumulative sum (CUSUM) approach is used to develop incident detection algorithms using loop detector data and probe vehicle data respectively, while the BP neural network combines the outputs from both incident detection algorithms. The proposed algorithm is tested and evaluated with the data generated by the simulation model INTEGRATION. The result shows that the outputs using BP neural network improves the accuracy provided by each single source incident detection algorithm.