Abstract
This paper presents a new automated incident detection framework for both freeways and urban arterial roads. A common modular architecture that includes a special data processing module to handle site specialties is applied to the freeway algorithm (TSC_fr) and the arterial road algorithm (TSC_ar). Bayesian networks are constructed to store general expert traffic knowledge and perform universal incident detection. The TSC_fr algorithm is evaluated using a large number of field incident data sets, and the TSC_ar algorithm is tested using simulation data. The testing results are very encouraging. It is found that both detection rate (DR) and false alarm rate (FAR) are not sensitive to incident decision thresholds. When the decision threshold is above the certain level, both DR and FAR reaches a very stable region. This is the unique feature of the TSC algorithms. The results also demonstrate algorithm transferability is achievable under the new incident detection framework.
