An important function that ITS software can provide for Road Network Operations is the ability to detect incidents and abnormal conditions from automatic analysis of the real-time traffic surveillance data. The development of Automatic Incident Detection (AID) algorithms began in the 1970s, and since then many algorithms have been developed. They have had mixed success, primarily because of their relatively high false-alarm rates (measured as the ratio of the number of false detections and the total number of observations).
AID algorithms can be broadly divided into four groups based on the principle behind the algorithm’s operation. These groups are:
These are among the most commonly-used AID algorithms. They are based on the premise that the occurrence of an incident results in an increase in the density of traffic upstream and a decrease in traffic density downstream. The California Algorithm is one of the earliest comparative-type AID algorithms to be developed – and is often used for comparisons and bench-marking. Since the original California algorithm was first developed refinements have been made to its performance. At least 10 new algorithms have been produced, of which algorithms 7 and 8 are the most successful. The TSC 7 algorithm represents an attempt to reduce the false-alarm rate of the original algorithm. The TSC 8 algorithm test repeatedly for congestion effects upstream of an likely incident and monitors other traffic characteristics.
Catastrophe Theory derives its name from sudden changes that take place in one variable that is being monitored – whilst related variables, also being monitored, show smooth and continuous changes. For incident detection, catastrophe theory algorithms monitor the three fundamental variables of traffic flow – namely speed, flow and lane occupancy (density). When the algorithm detects a drastic drop in speed, without an immediate corresponding change in occupancy and flow, this is an indication that an incident has probably occurred. The McMaster algorithm developed at McMaster University in Canada is a good example of an algorithm based on this concept.
Statistical and time series methods are used to forecast future traffic states or conditions. By comparing real-time observed traffic data with data forecasts, unexpected changes are classified as incidents. An example of these algorithms is the Auto-Regressive Integrated Moving-Average (ARIMA) time series algorithm. ARIMA is used to provide short-term forecasts of traffic occupancies based upon observed data from three previous time intervals. The algorithm also computes the 95% confidence interval. If observations fall outside the 95% range as predicted by the model, an incident is assumed to have occurred.
Several Artificial Intelligence (AI) concepts have been applied to problems in transport engineering and planning. Automatic incident detection is one application. Detecting incidents is a good example of a group of problems known as pattern recognition or classification problems – for which AI theories are quite effective in solving. Amongst the AI concepts most often applied to the problem of incident detection are Artificial Neural Networks (ANNs). These use complex algorithms and multiple computer processors to recognise patterns and connections in the input data.