An automated incident detection procedure developed for arterials detected 75 percent of reported incidents and had a false alarm rate of 26 percent.

Experience with a Bluetooth-based vehicle re-identification method in Oregon.

Date Posted

Arterial Incident Detection Procedure Utilizing Real-Time Vehicle Re-Identification Travel Time Data

Summary Information

Travel time data obtained from vehicle re-identification systems is becoming increasingly available due to the implementation of various technologies such as license plate recognition, inductive loop signature systems, and Bluetooth-based wireless vehicle identification. These advances present an opportunity to develop and apply automated incident detection methods. However, travel time data and incident detection have both been traditionally focused on freeways and other free-flowing roads. This research developed an incident detection procedure to better support travel time data for arterials. The accuracy of the procedure was evaluated by comparing the results with reported incident data.

Incident detection methods have primarily fallen into three categories: roadway-based algorithms using loop detector data, probe-based using data from vehicles equipped with toll transponders or GPS receivers, and driver-based techniques that identify driver responses to incidents. The method developed in this research is considered a probe-based approach. However, most probe-based incident detection relies on travel time data and so this research is unique in that it uses a new method (vehicle re-identification) to support incident detection on arterials. The objective of the incident detection procedure is to correctly identify potential incidents as soon as possible, and to do so with a low false alarm rate.


The researchers collected travel time data along an arterial in Oregon using a Bluetooth enabled vehicle re-identification system. Historical travel time data was also collected to establish thresholds for each day and each time period. The researchers compared the travel time samples and the time intervals between successive travel time samples with the pre-established thresholds for each day. A potential incident was indicated when a pre-set number of consecutive travel time samples all exceeded the threshold for that time period.
  • The researchers collected travel times for individual vehicles between five different intersections along an arterial in Tigard, Oregon (Highway 99W).
  • Also collected was the time between the generation of consecutive travel time samples.
  • Oregon Department of Transportation provided data on reported incidents on the same arterial and during the same time periods.
Data was collected using a Bluetooth-based vehicle re-identification system that had already been installed at five different intersections on the same arterial. Travel times between the data collection units (DCU) were collected continuously from June 9, 2011, through February 29, 2012, and from May 7, 2012, through May 29, 2012. The corresponding incident reports used for validating the accuracy of the procedure were provided by Oregon DOT.
  • Although travel time data was collected continuously 24 hours per day, the incident detection procedure was evaluated on travel time data collected from 6:00AM to 8:00PM (including weekends) when there was a consistent traffic flow.
  • This incident detection procedure focuses on identifying potential incidents, but as such cannot alone be used to distinguish between a real incident and an unusual event that would not be classified as an incident but might still have an impact on traffic flow.

The performance of the procedure was evaluated based on the detection rate and the false alarm rate.
  • Eight reported accidents occurred during the time periods for which travel time data was collected. The incident detection procedure detected six, resulting in a detection rate of 75 percent.
  • To evaluate the false alarm rate, the number of potential incidents detected was compared with the number of false alarms over a 2-month period. During the two months, 46 potential incidents were detected, and 12 were determined to be false alarms, resulting in a false alarm rate of 26 percent, or 6 false alarms per month. Compared to a national survey on maximum number of acceptable false alarms (10 false alarms per day), this false alarm rate is low.
Based on these findings, the researchers suggest that the procedure effectively balances incident detection rates and false alarm rates.
Goal Areas
Deployment Locations