Automatic Incident Detection on Freeways

The purpose of automatic incident detection algorithms is the identification of traffic disturbances. Urban and freeway networks require different types of methods due to the diversity of traffic flows.

The research topic at the department is observing the usability of freeway incident detection algorithms. Traditional logics check certain parameters of the traffic flow (usually in every 30 seconds) and compare them to fixed threshold values. If a value is beyond the threshold, the algorithm sends an alarm signal. Therefore, calibrating parameters of algorithms is a key issue. Too loose threshold values result in a high false alarm rate, whereas too strict thresholds lead to increasing number of undetected incidents.

An example of exceeding the threshold value (occupancy)

Former pieces of research focused on calibrating parameters of single algorithms. The research done at the department has two unique aspects:

- the threshold values of algorithms are not fixed values, they are functions of traffic volume, that has proven to be the most important factor of incident detection

- and at a specific section not only one but also more types of algorithms have been applied simultaneously. The value of traffic volume continously determines the algorithm type that should be used on a section.

Therefore the main goal of the research was determining appropriate types of algorithms and their threshold values among different traffic volume conditions. Three requirements have to be met: if an incident has occurred, it has to be detected but when no incident happens no alarm signal is accepted and certainly the time to detect has to be as low as possible.

The relationship among the measures of performance.

If one becomes better, another one becomes worse. This makes calibration difficult. The test network has been constructed in Vissim microscopic traffic simulator. The algorithms have been coded in MATLAB using the Vissim COM interface. This way the two pieces of software are connected. MATLAB logics observe the Vissim traffic state real-time and give alarm signals if needed.

Vissim-MATLAB real-time communication