This site will look much better in a browser that supports web standards, but it is accessible to any browser or Internet device.

ESGI49
29/03 - 02/04/04 OCIAM Oxford University
OCIAM

Home | About | Programme | Problems | Registration

Real time traffic monitoring using mobile phone data

Michael Alger, Vodafone Pilotentwicklung GmbH

pdf

Background

When someone makes a mobile phone call while travelling by road, the call has to be handed over from one Base Transceiver Station (BTS) to the next, and the timing of these handovers enables the vehicle speed to be estimated. We have extracted GSM signalling data from a selected area around Munich during three months last summer in order to detect road traffic congestion information directly from the mobile network. As a result, we obtained noisy velocity-over-time data,  neither equidistant in time nor exact.

Problems

The problems are:

  • Find a filter that, with the least possible delay, detects sharp declines of the average speed and hence tells us the beginning of a traffic jam.

  • Find a filter that detects the return to normal conditions after a traffic jam, also as soon as possible.

  • Find a filter to give a reasonable estimate for the possible speed at which one could expect to be able to travel by car during normal or congested conditions. (The speed distribution obtained from GSM data is the sum of the distribution of car speeds and that of truck speeds, and so is often bimodal.)

This is a sample of the data for illustration:

Each red dot represents one observed traveller who passed a given road segment with the corresponding speed at the given time. The black line is the result of a rudimentary sample filter, gliding average of the last 32 values. Data for Thursday, Friday and Saturday shows traffic congestion as the travelled speeds drop sharply. Note that during night time there is barely any data while on weekends there is less data than during the week. This should not be confused with standing traffic, however.

Note that the filter shall be used in near-real time while the data is generated. Hence it may not take future data values to compute a result for a given point in time.

The identification of such a filter may help to introduce different traffic information services. I will be glad to take comments and answer any questions.

The complete data spans a trial carried out during three months last summer on approximately 110km motorway around Munich, Germany. It is organized in a set of 46 excel-sheets and will be supplied on CD.

Home | Contact

This page last modified by C. Breward
Monday, 15-Mar-2004 10:57:27 GMT
Email corrections and comments to mortimer@maths.ox.ac.uk