Background

This research is part of the Intelligent Intersection project (see GRETIA- The Intelligent Intersection project: the management of signalized intersections ).

Using a multi-camera device linked with a scene analysis system, the aim is to detect relevant characteristics of the users' movements in relation to the traffic signals, in order to design and follow up a broad spectrum of indicators. Changes to these indicators can be studied at very short intervals (signal cycles) and analyzed according to the traffic conditions (demand magnitude, incident occurrence, etc). Thanks to a database of traffic scenes observed for various control strategies, it is possible to study the impact of traffic control on these indicators.

Objectives

The objectives of the study can be described at three different levels:

•  Methods : the aim is to enhance the assessment of traffic signal control strategies, by including the effects induced on the users' behavior and on the environment;

•  Results : these methods can be used to highlight the potential benefits of real-time control strategies, such as the CRONOS method developed at INRETS (see GRETIA - CRONOS: a real-time urban traffic control algorithm );

•  Operational level : an integrated system is being developed for the management of signalized intersections, which will allow multi-criteria assessment of the control strategy.

A transversal theme

Reconstitution of the traffic flow

The scene is observed by several cameras which give information about the road occupancy but don't give information about individual vehicles; we need to reconstitute the movements across the intersection and develop tools for the interpretation of video and loop sensor measurements.

The traffic flow is reconstituted at two complementary levels:

The whole scene : the first approach aims at reconstituting the space-time coherence of the flows crossing the whole scene. The system reconstitutes the origin-destination flow at each signal cycle: the demand observed for each input link and each movement (straight and turning).

•  The key areas of the intersection : the analysis is focused on the key areas of the intersection, be they sensitive or complex, such as stop-lines. One module of our system processes locally perceived patterns and classifies video measurements through self-organizing maps (a neural network model).

Artificial Intelligence approaches
Thanks to the structure of the Intelligent Intersection system, it is possible to develop various modules within a real-time object-programming environment, using several techniques from artificial intelligence: rule-based systems,time-dependent reasoning, neural networks, temporal database. The problems we are investigating belong to active research fields such as context influence in pattern recognition, permanent learning, focusing capacity, etc.

Three areas of expertise

Control strategy and traffic flow

The effects of the traffic signal control strategy on the distribution of vehicles on the intersection need to be quantified. Each cycle of a given signal (green/amber/red) has to be characterized according to the induced presence of vehicles in front of the traffic signal and its conflicting signal: time spent, delay, queue length, etc. The distribution of these various characteristics is analyzed and compared for various control strategies when demand is similar. The aim is to find discriminating features between strategies, and on that basis to design new indicators to measure the efficiency of a control method.

Control strategy, comfort and safety

We are investigating red light running and the effect of the control strategy on violations. Focusing methods are being used to analyze signal stop-lines.

Assessing the effects of the control method on safety does not only concern violations. A PhD student is currently doing research into the effect of control strategy on behavior and risk. The method of traffic conflicts should identify a continuum of situations indicating discomfort or risk. These situations will be detected automatically. Some new indicators of the efficiency of the traffic signal control strategy will also be used.

Control strategy and pollution

An assessment model of the environmental costs induced by an isolated signalized intersection has been developed in order to estimate average pollutant emission and fuel consumption for different categories of vehicle (diesel, gasoline with or without catalyst converter). The model estimates the average CO 2 emissions linked to the greenhouse effect, and other pollutants (CO, HC and NO x ).

The vehicles contribute in different ways to the calculated assessments depending on whether or not they stop at signals; work on the reconstitution of the whole traffic scene is used to quantify the number of vehicles that did or did not stop per movement.

The model has been calibrated using real urban driving cycles and corresponding instantaneous emission measurements provided by the INRETS-LTE laboratory. It has enabled us to quantify the consumption/pollution costs for each traffic signal control strategy.

References

[1] Midenet S., Boillot F., Pierrelée J.C., "Signalized intersection with real-time adaptive control: on-field assessment of CO 2 and pollutant emission reduction" , Transportation Research Part D, in press, 2003.

[2] Saunier N., Midenet S., Grumbach A., "Automatic detection of mobile interactions in a signalized intersection" , 16 th ICTCT Workshop, Soesterberg, The Netherlands, 29 th October - 1 st November 2003.

[3] Boillot F., Midenet S., Pierrelée J.C., "Real-life Cronos evaluation", 10 th  International Conference on Road Traffic Information and Control, IEE London England, n°472, pp 182-186, April 2000.