Objective
The main objective of the project is to develop an integrated congestion and incident detection system which will incorporate computer vision techniques. This system will be designed to be a sub-system of a wider traffic control system to be developed by other DRIVE projects. The module developed by this project will have the option of using image processing either centrally or locally.
The INVAID system is an automatic traffic incident detection system for both motorway and urban network applications which uses computer vision techniques. The system involves the automatic digitising, processing and interpretation of pictures from roadside closed circuit television cameras. INVAID incorporates 2 of the leading methodologies specifically designed for the analysis of live road traffic images at two levels: IMPACTS (UCL) and TITAN (INRETS). The INVAID project has designed algorithms that perform automatic incident detection based on data output using these two techniques at both individual sensor (ie television camera) level and at system wide, multiple camera level.
The objective of the central system (CS) is to collect information transferred by highway surveillance cameras in order to provide a spatial instantaneous description of the state of the traffic, as well as the temporal evolution of incidents and traffic conditions detected by sensors and to make this information available to the operator. This interpretation offers a potential for detecting both directly visible incidents within the real camera road segment (RCS) and for detecting abnormalities that may be the result of an incident elsewhere (ie in a road segment not covered by cameras, the virtual camera road segment (VCS)). The main internal functional modules are:
a communication interface to provide the link between the CS and all installed local sensor modules (LSM) and data collection services;
a decision making process for short and long term processing of incoming data;
a man machine interface module (MMI) located at the top level of the system hierarchy to make all the system's capabilities easily accessible and controllable by the user. The MMI provides a graphical environment in which a great amount of information can be displayed to the user.
The local sensor module (LSM) performs the image processing and computer vision procedures necessary for automatic incident detection. Each LSM is attached to one video camera. Its main task is to provide alarms for incidents occurring in the camera field of view or nearby. There are 2 types of sensors called Type A and Type B.
The Type A sensor is based on a TRISTAR like image processing module and it was developed by INRETS. It primarily aims at detecting incidents that occur within or nearby the camera field of view (usually, several hundred meters of motorway are covered by one camera). It does not produce only a binary alarm but provides a more detailed information such as there is an incident of type T or the confidence level of this alarm is C. In addition, the local system acts as a traffic sensor and calculates a set of traffic measurements such as volumes, velocities and concentration. Other optional functions of the system (eg communication, image compression and transmission) are related more to industrial problems rather than to research.
The Type B processor is based upon a particular computer vision technique, IMPACTS, researched and developed at University College London. The incoming surveillance camera images are processed to give a spatial description of some of the basic traffic characteristics. This description is of a qualitative nature and aims to reflect the behaviour of the traffic directly. It does not produce the classical measures of traffic flow, detector occupancy and speed which are found by aggregating data from vehicle detection. The spatial description output from the Type B algorithm shows where within the image of the road there is moving, stationary and no traffic. Information is also sought from existing conventional detectors and from a small database. By combining this information, robustness of the incident detection should be improved.
The main objective of the research, entitled project INVAID, was to develop an automatic incident detection system for both, motorway and urban network application, using computer vision techniques. Others objectives were:
to identify and produce a standard classification of incident types for traffic behaviour on motorway and urban roads;
to develop suitable techniques that will combine computer vision, existing traffic data and other relevant data to provide a robust system;
to design a man machine interface that will allow a traffic controller to rapidly understand the data presented by the system and take the appropriate action.
The INVAID system uses computer vision techniques and is primarily concerned with the interpretation of the image processorsinformation and the preparation of information to be sent to operators. The system comprises 3 processing modules to detect and interpret incidents:
2 local sensor modules (LSM) that takes input from the cameras situated at road side sites;
the central system module that takes the input from the LSMs and additional information from other ATT systems.
The principal elements of the INVAID Type B processing module for automatic traffic incident detection on high speed roads is described. High speed roads are dual carriageways, multilane highways where stopping is only permitted in emergencies, or only occurs as a direct consequence of an emergency or congestion. Permitted traffic speeds are normally greater than 100 kilometres per hour and intersections are grade separated.
The Type B processor is based upon a particular computer vision technique, IMPACTS, researched and developed at University College London. The incoming surveillance camera images are processed to give a spatial description of some of the basic traffic characteristics. This description is of a qualitative nature and aims to reflect the behaviour of the traffic directly. It does not produce the classical measures of traffic flow, detector occupancy and speed which are found by aggregating data from vehicle detection.
The Type B processor is envisaged as being located at a control centre, with video images being transmitted from remote cameras. Multiple cameras would each have their own associated Type B processor which would then be fed into the central system.
The techniques developed were implemented in the laboratory on a SUN workstation based system. The Type B algorithms run entirely on software in the SUN and have been written in C. Several trial runs have been carried out using real time video recordings of various high speed road scenes, some of which included incidents. In one example the traffic goes from free flowing to congested, and then an accident occurs at the far end of the view. The Type B output, shown as a text string, correctly logs this sequence of events.
The main objective of the research was to develop and automatic incident detection system for both, motorway and urban network application, using computer vision techniques. Computer vision involves the automatic digitizing, processing and interpretation of pictures from roadside closed circuit television (CCTV) cameras. The system developed comprises 3 main modules:
2 data/image processing modules, each using different camera arrangements, image processing and algorithms;
the central system which takes its input from the data/image processing modules and from other road transport informatics (RTI) systems.
The automatic incident detection system has the following advantages:
the road area observed is increased from 2 to 3 metres to 200 to 300 metres;
an artificial intelligence (AI) approach broadens the range of traffic conditions over which effective incident detection can be performed;
external information from existing sensor systems about the nature of the highway and temporary changes, such as roadworks, can be incorporated into the rule based approach allowing a greater flexibility of used;
the information returned by the basic sensing device, a video camera, is readily understood by human traffic controllers.
The principal objectives of this project can be summarised as follows:
- to identify and produce a standard classification of incident types motorway and urban networks,
- to utilise computer vision techniques to provide new spatial measures of traffic behaviour on motorways and urban roads,
- to develop artificial intelligence techniques that will combine computer vision, existing traffic data and other relevant data to provide a robust automatic incident detection system for both motorway and urban networks,
- to design a man-machine interface that will allow a traffic controller to rapidly assimilate the data presented by the system and take the appropriate remedial action,
- to monitor the development of other RTI systems and specify interfaces between them and this system. This will enable the system under development in this project to be incorporated into wider traffic control systems,
- to evaluate the performance of the system in field trials.
Main Deliverables:
Prototype for field trial on urban and motorway sites.
Fields of science
- engineering and technologyelectrical engineering, electronic engineering, information engineeringelectronic engineeringcontrol systems
- natural sciencescomputer and information sciencesdatabases
- engineering and technologyelectrical engineering, electronic engineering, information engineeringelectronic engineeringsensorsoptical sensors
- natural sciencescomputer and information sciencesartificial intelligencecomputer vision
- social sciencessocial geographytransport
Topic(s)
Data not availableCall for proposal
Data not availableFunding Scheme
Data not availableCoordinator
Spain