Traffic congestion has a dominant effect on vehicle fuel consumption. This project will identify how congestion data would support despatch and routing strategies. It will also develop traffic models based on enhanced traffic monitoring methods, primarily from video camera input data. These models will predict statistical road usage data for time-of-day and day-of-the week on all main roads in a given area. The methods will enable typical journeys to be known in sufficient detail so that unusual flows at specific locations can be converted into predictions of road usage levels. Likely locations for congestion will be predicted before the congestion actually occurs. Follow-on projects will use this information to develop methods to optimise freight vehicle routing so as to minimise fuel consumption.
Current traffic monitoring systems measure average vehicle speeds on roads by a variety of techniques to detect when congestion is occurring. However they are not able to predict vehicle flow rates over the coming few hours and hence have only limited ability to redirect traffic in ways to minimise overall congestion of the roads. This project will identify the value of and develop enhanced traffic monitoring methods, primarily from video camera input data, that will automatically develop statistical road usage models for time-of-day and day-of-the week on all main roads in a given area. The method will enable typical journeys to be known in sufficient detail, so that abnormal rates of departure from specific locations can be converted into modifications to journey patterns over the coming hour or two. Then road usage levels over a similar timescale and hence likely locations for congestion can be predicted before congestion actually occurs. Prediction of congestion in advance will allow much more effective congestion avoidance and congestion mitigation strategies to be employed. This project will explore and quantify the value of such strategies using the new capabilities in congestion prediction.
The technology novelty of the project lies in the improved data gathering capabilities of new image understanding systems, such that a camera at a junction can analyse a scene as effectively as a human observer, 24/7 directly into the main database, at a tiny fraction of the cost of a human. Such vision performance has only recently become achievable at modest cost due to recent developments in image processing hardware and software (e.g. GPU hardware systems and efficient multi-resolution wavelet-based decomposition methods, with advanced machine learning methods for very large databases) [1-3]. It should be emphasised that the proposed image understanding approach has many advantages over automated number-plate recognition (ANPR) technology for this application because it can be independent of DVLC, does not infringe driver privacy, and allows wide-angle views of the junction to be used (minimising the cost of camera hardware).
(i) To develop vision systems that are optimised for traffic monitoring at road junctions, so that each vehicle using the junction can be logged for vehicle type, route and speed through the junction, distinguishing marks (e.g. colour and logos), and any other relevant parameters.
(ii) To develop an analysis system which can integrate this data into journey tracks via many junctions and provide raw journey data to the main traffic database.
(iii) To develop a statistical model for road use in a region, based on the journey tracks.
(iv) To develop a congestion prediction system for the above model and apply this to optimization of vehicle routing
Academic impact: Insights into how congestion prediction might impact scheduling and routing decisions will create a spur for new research in these specialist domains. Creating the vision system for monitoring traffic, able to provide robust discrimination under most weather conditions, day and night, will provide a powerful tool for researchers across many other disciplines. The design of the database poses novel general problems of size, complexity and robustness to significant error rates in the input data. Solving these technical challenges will provide powerful resources for others.
Commercial and social impact: Using real time congestion data and prediction might have the potential to save considerable time, cost and frustration. Measuring traffic flows and vehicle classes at low cost, across many more road junctions than current systems, could revolutionise the accuracy of traffic information systems. The transition from flow monitoring to congestion prediction ability is pivotal. All road users will benefit from the information to complete their journeys with minimal disruption, from substantial gains in fuel efficiency and carbon reduction, lower costs to all road users, and a significant drop in frustration levels for all drivers. These benefits will be vital given the forecast increases in future road usage.
- Selesnick, I.W., R.G. Baraniuk, and N.G. Kingsbury, The Dual-Tree Complex Wavelet Transform. IEEE Signal Processing Magazine, 2005. 22(6): p. 123-151.
- Nelson, J.D.B. and N.G. Kingsbury, Enhanced shift and scale tolerance for rotation invariant polar matching with dual-tree wavelets. IEEE Trans on Image Processing, 2011. 20(3): p. 814-821.
- 3. Bendale, P., T. W., and N.G. Kingsbury, Multiscale keypoint analysis based on complex wavelets, in British Machine Vision Conference 2010, 31 Aug – 3 Sept,. 2010: Aberystwyth, Wales.