An Evaluation of Existing Incident Detection Algorithms (1232-20)

Several algorithms have been proposed for automatically detecting incidents using freeway detector data. Some are comparative algorithms that compare measured traffic conditions to preestablished thresholds. Other use statistical procedures to detect significant changes in traffic patterns over time. Still others use complex theoretical models to predict future traffic conditions using current traffic measures and historical trends. The structure of an algorithm affects its performance in terms of detection rate, false alarm rate, and detection time.|No single algorithm appears to be superior in terms of its reported performance, data requirements, ease of implementation, ease of calibration, and operational experience. Using the results from both off-line and on-line evaluation reported in the literature, most algorithms appear to achieve the same level of detection and produce similar false alarms rates, when properly calibrated. Detection times for most algorithms are also similar.|Given the results of the evaluation and the sites visits, it is recommended that TxDOT consider the following incident detection algorithms in the initial implementation of their freeway surveillance and control centers:|Modified California Algorithm #7|Modified California Algorithm #8|McMaster Algorithm