Urban Traffic State Estimation Using Probe Vehicles

Abstract

An Intelligent Transportation System (ITS) is a system which uses traffic information to generate useful information about the congestion of road network with the aim of proper utilization of the road network. Traditional techniques to get traffic information relies on fixed road side sensors, inductive loop detectors and video cameras etc. This technique however is not scalable because of high cost of such equipment to be installed on a large road network. An alternate to this technique is the use of Global Positioning System (GPS)-based probe vehicle data. This technique is scalable to a large road network due to the presence of such probe vehicles (taxis). This data can be collected by the use of smart phones. However, this technique faces the challenge that the data collected through this medium is sparse and unevenly distributed due to random movement of the vehicles throughout the road network. So this kind of data can lead to inaccurate traffic state prediction results. To overcome this challenge, various techniques have been proposed to measure traffic state through this kind of data for which this study provides a literature review along with the research gaps. In this thesis a model has been proposed to measure the traffic state in the form of average speed of vehicles on a road. The main focus of this research work is to gather the traffic data and present the state of data in an efficient way using Recursive Partitioning and Regression Trees (RPART) machine learning model so that this information can be used for other future developments like urban infrastructure planning, re-routing of traffic etc. The data used for training and testing the proposed model comes from the simulation of an Indian city road network by the use of traffic simulation software Simulation of Urban MObility (SUMO) and is pre-processed before using it to train the proposed model.