Transferability of freeway incident detection algorithms

This research seeks to evaluate the potential for transferability of new freeway incident detection algorithms that can be used in automatic incident detection. An incident detection logic is evaluated that distinguishes incidents from recurrent congestion and other traffic disturbances using exponential smoothing. The algorithm is tested with loop detector data from test sites on I-35W in Minnesota and I-880 in California with promising results. Test results indicate that, at the Minnesota test site, algorithm performance improves over earlier findings with a limited data set, without requiring recalibration. At a detection rate of approximately 60 percent, the algorithm produces one false alarm per hour in the total (southbound and northbound) 40-km (25-mi) test section, which includes 54 detector stations. When the algorithm is recalibrated and tested at the California site, its performance further improves. The performance of the algorithm is compared with that of major algorithms of comparative type and is found to be superior at all times. This superiority is not affected by the transport of the algorithm across sites; it is still observed in analyzing northbound and southbound, peak and off-peak, and accident versus nonaccident detection. The findings indicate a strong transferability potential of the new algorithm.