An evaluation of freeway incident detection algorithms using field data

Effective incident detection and management on freeways is vital in order to maximise road system performance and minimise the problems associated with growing traffic congestion. A comparative study of freeway incident detection algorithms was undertaken on the California algorithm, the University of California, Berkeley (UCB) algorithm, the ARRB-VicRoads algorithm, the Detection Logic with Smoothing (DELOS) algorithm, and an artificial neural network (ANN) model. It was found that the ANN model performs better than the other rule-based algorithms. The California and DELOS algorithms performed the best out of the four rule-based algorithms that were evaluated. It is important to note that training an ANN model is far more complex than calibrating a rule-based algorithm. The ratio of incident to non-incident data in training data sets can be critical to the success of the ANN model. On the other hand, the calibration of rule-based algorithms is more straight forward. An optimisation software FRIO has also been developed to optimise the calibration and, thus, maximise the performance of rule-based algorithms.

Language

  • English

Media Info

  • Pagination: 459-76
  • Serial:
    • Volume: 23
    • Issue Number: Part 1

Subject/Index Terms

Filing Info

  • Accession Number: 01394356
  • Record Type: Publication
  • Source Agency: ARRB
  • ISBN: 0730724905
  • Files: ATRI
  • Created Date: Aug 23 2012 11:39AM