Citation

Levin M, Krause GM. Transp. Res. Rec. 1979; 722: 49-58.

Copyright

(Copyright © 1979, Transportation Research Board, National Research Council, National Academy of Sciences USA, Publisher SAGE Publishing)

DOI

unavailable

PMID

unavailable

Abstract

ive incident-detection algorithms of the pattern-recognition type were evaluated off-line by using incident and incident-free data collected on Chicago's expressways under various traffic and environmental conditions. Algorithm efficiency was evaluated in terms of detection and false-alarm rates and mean-time-to-detect. Evaluated were a comparative analysis of algorthm efficiency, the efect of lateral detechorization on algorithm performance, a hierarchical analysis of threshold effectiveness and the effect of incident severity on algorithm performance. Although no specific algorithm was found to be superior for levels of detection lower than 95 percent, for higher levels of detection one algorithm developed by technology Services Corporation was found to be best. The algorithms did not differ statistically in mean-time-to-detect, which ranged from 2 to 4 min, rendering this parameter ineffective in algorithm selection. The relation between detection rate and false-alarm rate, however, was found to be the critical criterion for algorithm selection. Feature thresholds developed for detector-lane incidents were found to be less sensitive to traffic-flow disturbances than were thresholds developed for non-detector-lane incidents, thus yielding lower false-alarm rates. Analysis of algorithm performance under various traffic and environmental conditions revealed that thresholds developed for a representative sample of incidents were effective when used on the "rush wet", "nonrush dry", and "nonrush wet" traffic data. Therefore, less effort was needed to develop the set of thresholds. Thresholds developed for accidents occurring on the detector lane proved to be effective in detecting accuring on the detector lane proved to be effective in detecting accidents and nonaccident incidents on both the detector and nondetector lanes.