Reliability improvement in automated incident detection (AID)

Abstract

This study uses the simulated data collected from the probe vehicles and loop detectors to explain how the Adaptive Neuro-Fuzzy Inference System has been developed to be applicable in the Automatic Incident Detection on the arterial roads. This research is conducted to extend what previously have been done in this area of study, and it is theoretically built on those findings that support the effectiveness of the Adaptive Neuro-Fuzzy Inference System in the data fusion. Because it is difficult to collect real data from the road networks, in this study, we use a data set formed by a validated and calibrated traffic simulation model of a commuter corridor located in Brisbane, Australia. Simulated accidents were provided and the required data were gathered from the probe vehicles and loop detectors that have been deployed at two different places of the network. A detector configuration was examined, and a total number of 108 incidents were modelled for that. To ensure the generality, the models were differed in factors such as the incident location, incident duration, road and detector configuration, severity level of the incident and the traffic flow conditions. The best result that was obtained for the Adaptive Neuro-Fuzzy Inference System was a 95% detection rate for a false alarm rate of 0.5%. The data collected for this study were consisted of features like speed, occupancy, and flow.