Traffic patterns become more complex as a road network grows to meet demand or improve access. Higher levels of congestion on some parts of the road network mean that unplanned incidents can have an influence outside the immediate area of the incident. This can result in greater risk to human safety and economic loss. (See Traffic Incidents)
Automatic Incident Detection (AID) use computers to continuously monitor traffic conditions and detect incidents or traffic queues. Road networks equipped with a variety of detectors and CCTV cameras provide the basis for automated continuous monitoring. The detectors are deployed to provide speed, occupancy and flow data to assist with traffic operations. Their output can be used to alert control centre operators to unusual traffic conditions and to detect incidents much more quickly than by manual methods – and so manage them more effectively.
It has been estimated that for every minute saved in detecting incidents, verifying them, and clearing them – another five minutes are saved in recovery time. The use of AID can also improve road safety by enabling a more rapid response to the victims of accidents within the critical first hour, otherwise known by the emergency services as the ‘golden hour’.
Incident management involves the implementation of a systematic, planned and coordinated set of response actions and deployment of resources to prevent accidents in potentially dangerous situations and to handle incidents safely and quickly. They are a critical component of a decision support system for traffic operations on the urban and interurban road network and its infrastructure of bridges and tunnels. (See Incident Response Plans)
At the heart of Automatic Incident Detection (AID) are filters and algorithms programmed to scan and automatically recognise a variety of patterns. The operators can validate the incident using conventional methods such as CCTV cameras and information from other sources, such as reports by the police and emergency services at the scene of the incident.
AID systems use the available detector data and image streams from road sensors and CCTV cameras to continuously monitor traffic conditions for patterns that indicate that an incident may have occurred. When the system detects a potential incident, it can alert traffic operators to commence the Incident Management (IM) process and implement pre-programmed responses to warn drivers and others. AID provides an effective means of improving incident detection. It complements manual monitoring of a road network since it can detect incidents quickly (often within a few seconds) and is able to monitor more video streams than a manual operator.
The most common sources of primary data for AID are in-vehicle inductive loops and CCTV cameras although there are a large range of other technologies that may be used. (See Vehicle Detection) The equipment (sensors and CCTV cameras) is used for both incident detection and the daily management of the network.
AID depends on the interpretation of information (vehicle occupancy, speed and traffic volume) captured from the network of traffic detectors. The Incident Detection Algorithm (IDA), often used in combination with AID, should be responsive and accurate - with a low error rate - to ensure that the AID system is trusted. Typical IDAs that depend on a network of discrete traffic sensors include:
Whatever algorithms are chosen, they will need to be calibrated and matched to the available types of sensors. It can be advantageous to simultaneously use multiple IDAs that have complementary characteristics – or apply different IDAs to different scenarios. To ensure accuracy, the data from individual sensors needs to be filtered before it is used by the IDA. (See Data Aggregation and Analysis) When an incident is detected, operators can draw on additional information from other sources such as CCTV cameras – and apply their own judgement to manually validate the incident.
IDAs can be applied to existing video streams. The IDA is either integrated within individual CCTV cameras or applied to a group of cameras to permit discrete events to be detected, such as stopped vehicles – as in the image below, debris, and vehicles driving erratically or in the wrong direction. Video-based AID detects events visually anywhere within an image - not only in the driving lanes but also on hard shoulders and emergency refuge areas regardless of the road surface material (concrete, asphalt, gravel or metal).
AID depends on there being comprehensive coverage by sensors and/or CCTV cameras of the road network to be monitored. The accuracy and geographical reach of AID can be improved if additional sensors and cameras are deployed.
Traffic sensors installed for surveying purposes or for general traffic monitoring may not be suitable for AID for various reasons (loop systems may detect traffic presence but not changes in traffic speed, existing CCTV cameras may not have image processing capability). In most cases additional sensors will be needed to ensure the reliable detection of incidents. (See Vehicle Detection)
Images taken from a CCTV camera dedicated to AID can be used to provide reliable information on occupancy, flow and speed - whilst also serving its primary purpose of surveillance - enabling an operator to validate an incident reported by the IDA.
For night-time conditions or where streetlights are unreliable, thermal cameras are effective at detecting vehicles on verges, or pedestrians and animals on roads. Thermal cameras also perform better in foggy conditions - so are useful for simple traffic monitoring.
Whether existing or new cameras are deployed, pre-recorded video of incidents can be used to test the performance of an AID system under different conditions, typically described by three parameters:
If hard shoulders are designated as normal travel lanes on a part-time basis (“hard shoulder running”) then an AID system, as part of a Decision Support System (DSS), is necessary to ensure the safety of road users. For example, if an incident is detected on the hard shoulder during hard shoulder running, approaching vehicles need to be alerted and the hard shoulder closed automatically or under manual supervision.
Excess traffic demand may cause a breakdown in traffic flow that results in the formation of queues and increases the risk of vehicle collisions and other incidents. An IDA function needs to be able to detect unusual queues, compared with normal traffic flows – adjusted for time of day, the day of the week and holiday periods.
Legacy networks of CCTV cameras can support AID - although a video decoder will be needed to prepare the video for computer analysis. Depending on the manufacturer, many legacy CCTV cameras, including those with a PTZ function, can provide video to support an AID function. It is likely that existing cameras have been positioned for surveillance – and may not be suitable for the image processing required by camera-based AID. They will need to be repositioned to ensure high detection accuracy and to remove the effects of light reflection (from vehicle headlights, wet road surfaces, low-level sunlight).