Transferability of Expressway Incident Detection Algorithms to Singapore and Melbourne

Abstract

This study investigates the performance of several existing automatic incident detection algorithms along the Central Expressway in Singapore and freeways in Melbourne, Australia. These algorithms were originally developed for freeways in the United States. Thus, it is of interest to evaluate how they would perform when applied to cities in other countries. The evaluation is carried out on two databases containing 160 and 100 incidents collected from Singapore and Melbourne, respectively. These databases reflect differences in vehicle detector system used to collect traffic parameters and in driver behavior. The following empirical findings were obtained: (1) the Minnesota and Standard Normal Deviate (SND) algorithms appear to possess transferable properties as well as being able to receive wide-area traffic measurements from a machine-vision vehicle detector system; (2) California Algorithm number 7 and the Double Exponential Smoothing algorithm performed poorly in Singapore but may be transferable to Melbourne freeways; and (3) no significant difference in average efficiencies between the better-performing algorithms (SND and Minnesota) on both databases.