Freeway incident detection based upon pattern classification

The detection of accidents and other capacity reducing incidents, e.g., occurrences of disabled vehicles in traveled lanes, on urban freeways is an important aspect of freeway traffic management. This function has been automated in several existing freeway surveillance and control systems in the form of incident detection algorithms. Multiple feature incident detection algorithms use two or more functions of traffic data and associated thresholds to signal the occurrence of incidents. These algorithms are constructed to distinguish patterns of traffic conditions distinctive of incidents. In this paper, a general approach to the calibration and evaluation of multiple-feature algorithms which are structured as decision trees is described. This methodology is applied to the California algorithm.