Urban Traffic Flow Prediction by Using GPS and Neural Networks Theory

Alexandrina Meneses, and Isabel Osório

Abstract:	In urban traffic networks there are “main roads” – usually the largest ones, with best pavement maintenance – familiar to many people, but, there are also alternatives, allowing the drivers to avoid traffic lights, pedestrian crossings, speed limits and so on. In order to recognize traffic patterns and, also, for traffic jam prediction, an experiment, with a GPS receiver in a car recording its trajectory, was implemented in a middle town, Santa Maria da Feira, 30 km south of Porto. A set of drivers of different ages, different cultural levels, different jobs and different knowledge about the town was selected. These persons were invited, several times, to drive under different weather conditions, at different hour of the day, all days of the week, from one same point, in downtown, to a shopping center, 5 km away, in the periphery. This shopping center is used not only by local people but also by people from the surroundings. On the top of the car was placed the external antenna of a TRIMBLE GEO 3 receiver and code measurements, with line feature, were recorded during all the travels. Two hundred eighty one arcs of the traffic network were available and the driver was free to choose the trajectory between the two points. He was only asked to fill a form with the name, the date and hour of the day and the weather conditions. The experiment took place during the months of January and February. All the trajectories were differentially corrected by using data from a reference station located at the City Hall. Afterwards, the solution was exported to the city GIS for the identification of the arcs of each trajectory. A data base of different entries was, then, created. These entries are: • Characteristics of all available arcs, such as, type of pavement and its state of maintenance, presence of schools, bus stops, pedestrian crossings, traffic lights and other possible conflicts with pedestrians; • Details about the drivers, such as, age, job and knowledge of the town; • Day of the week and hour of the day for each trajectory; • Weather conditions during the trajectory; • Passage or not by the driver on each arc during each trajectory. By using different combinations of these parameters as input, different Neural Networks were trained in order to detect traffic patterns and to predict the percentage of people using the same arc of the traffic network, under the same weather conditions, at the same hour of the day. In this paper, the data acquisition and processing are presented and the promising results obtained with different algorithms are discussed and, finally, the implementation of the experiment as a systematic process in some type of cars belonging to the City Administration is considered.
Published in:	Proceedings of the 2004 National Technical Meeting of The Institute of Navigation January 26 - 28, 2004 The Catamaran Resort Hotel San Diego, CA
Pages:	974 - 981
Cite this article:	Meneses, Alexandrina, Osório, Isabel, "Urban Traffic Flow Prediction by Using GPS and Neural Networks Theory," Proceedings of the 2004 National Technical Meeting of The Institute of Navigation, San Diego, CA, January 2004, pp. 974-981.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In