Incident Detection Algorithms for COMPASS - An Advanced Traffic Management System

Paper #:
  • 912767

Published:
  • 1991-10-01
Citation:
Masters, P., Lam, J., and Wong, K., "Incident Detection Algorithms for COMPASS - An Advanced Traffic Management System," SAE Technical Paper 912767, 1991, https://doi.org/10.4271/912767.
Pages:
16
Abstract:
:Advanced Traffic Management Systems (ATMS) provide the means for local transportation officials to monitor traffic conditions, adjust traffic operations, and respond to accidents. By providing early traffic incident detection and management, and by redistributing traffic to less congested portions of the highway network, ATMS can influence vehicle operators' route choices. COMPASS, a state-of-the-art advancedtraffic managementsystemimplemented in the Metropolitan Toronto area, has adopted most of the Intelligent Vehicle-Highway Systems (IVHS) technologies.This paper describes the logic and implementation of the automatic incident detection for COMPASS. The purpose of incident detection is to identify the potential occurrence of incidents in a traffic stream by analyzing the flow characteristicsof the traffic stream. The output of the incident detection function will form the basis for incident verification by the operator and implementation of traffic response plans. Two incident detection algorithms have been developed for the system, namely the All Purpose Incident Detection (APID) algorithm and the Double Exponential Smoothing (DES) algorithm.The APID algorithm is based on the California incident detection algorithms which have the general structure of a binary decision tree. The algorithm has been designed to handle different traffic patterns. For example, the light/medium traffic incident detection routines are more suitable for detecting incidents at light/medium traffic conditions than the general incident detection routine. Furthermore, the false alarm rate may be reduced by introducing the compression wave test and persistence test.The DES algorithm makes use of a short-term forecasting technique for detecting irregularities of a traffic variable (e.g. volume) in a time series. Atracking signal is obtained by dividing the cumulative error of a traffic variable (e.g. volume) by the current standard deviation of the same variable. An incident will be identified when the tracking signal deviates significantly from a pre-defined threshold value. The traffic variables currently defined for COMPASS are volume, occupancy and speed. The false alarm rate will be reduced if more tracking signals are used (i.e. with more traffic variables defined).The execution cycle for the incident detection algorithms can be any multiple of the raw traffic data gathering cycle (20 seconds for COMPASS), up to a maximum of nine. Moreover, the traffic data used for the APID algorithm can be averaged over a user definable period from one raw traffic data gathering cycle to a maximum of five minutes. However, data required for the DES algorithm can only be averaged over the execution cycle, due to the nature of the algorithm.The COMPASS system allows concurrent execution of three algorithms at the same time, but there is virtually no limit regarding the number of algorithms installed in the system. An incident will be declared based on a pre-defined logic combination of the three running algorithms (e.g. [algorithms A and B] or [algorithms [A or B] and C]). This logic combination can be changed by the user in real-time to allow total flexibility.Before the algorithms were implemented on the COMPASS system, extensive simulation was performed in order to prove the logic of the algorithms and derive a preliminary set of parameters, using the historical data from the Burlington Skyway in Ontario. At the same time the incident detection rate and false alarm rate were thoroughly examined.
Access
Now
SAE MOBILUS Subscriber? You may already have access.
Buy
Select
Price
List
Download
$27.00
Mail
$27.00
Members save up to 40% off list price.
Share
HTML for Linking to Page
Page URL

Related Items

Technical Paper / Journal Article
2010-04-12
Standard
2002-01-23