Comparative Performance of Freeway Automated Incident Detection Algorithms

Abstract

Common measures of performance of incident detection algorithms are detection rate, false alarm rate and mean time-todetect. These measures are not independent and it is therefore necessary to determine the underlying performance trade off. In this paper, the performance of the incident detection algorithm currently implemented on Melbourne’s freeways is evaluated based on a set of one hundred incidents that occurred on Melbourne’s freeways under varying traffic conditions. The results are interpreted in relation to the broader operational experience with the incident detection algorithm. An improved algorithm, based on artificial neural networks, is also presented. An independent set of forty incidents, not used in the development of either model, was used for comparing the performance of the two algorithms. Evaluation results, in terms of detection rate, false alarm rate and mean time-to-detect are presented using performance envelope curves that show the trade off in performance between the two models. The results clearly demonstrate the substantial improvement in incident detection performance obtained by the ANN model over the ARRB/VicRoads model.