Improved freeway incident detection using neural network based on pulse data of the loop detector

Abstract

This study attempts to develop a new freeway incident detection algorithm that uses the data of pulse lengths and pulse gaps from the loop detectors as parameters and apply LVQ neural network to process the data to determine if an incident occurs. This algorithm reduces greatly incident detection time, so it offers a reliable basis to rapidly process the traffic incidents. Meanwhile, the algorithm can make use of the self-learning ability of neural network to determine the different thresholds for various freeways. At last, as the simulation results shown, the new algorithm for incident detection has a lower false alarm rate(about 0.41%), a faster detection speed and a higher detection rate(about 97%). It's found to be potentially applicable in practice.