Crowdsourcing Real-Time Traveler Information Systems

In the last decade, the concept of collecting traffic data using location aware and data enabled smartphones in place of traditional sensor networks has received much attention. With a steady market growth for smartphones enabled with GPS chipsets, the potential of this technology is enormous. This combined with the pervasion of participatory paradigms such as crowdsourcing wherein individuals with portable sensors instead of physical networks serve as sensors providing information. Crowd sensed data overcome a number of issues with traditional physical sensor networks by providing wider coverage, real-time data, data redundancy and cost effectiveness to name a few. While there has been a lot of work on actual implementations of crowd sensed traffic monitoring programs, there is limited work on assessing the quality, and validity of crowd sensed data. A systematic analysis of quality and validity is needed before this paradigm can be more commonly adopted for traffic monitoring applications. To this end, research is underway to deploy a crowdsourced platform for monitoring and providing real-time transit information for shuttles that serve the University of Connecticut. The thesis develops a framework and an open-source prototype system that is able to produce real-time traveler information based on crowdsourced data. In order to build the prototype, first it implements a robust Hidden Markov Model based map-matching algorithm to position the crowdsourced data on the underlying road network and retrieve the likely path. The accuracy of the map-matching algorithm has been found satisfactory for the current usage even when the GPS points are sampled at low frequency. Next, to predict the travel condition across the network from the crowdsourced data, a travel time prediction algorithm, based on Regularized Least Square Regression, has been implemented as well. This travel time prediction algorithm, together with the map-matching algorithm, has been applied in a simulated crowdsourcing environment. The travel time prediction results of the simulation show that the prototype system is quite capable of predicting travel time even when the crowdsourced real-time data is sparse. The simulation tests the performance of the travel time prediction algorithm in different scenarios. From the demonstrated predictive performance of the implemented prototype system, this approach to providing real-time traveler information is found promising. It is also possible to apply the prototype to all regions and all modes of transportation, exploiting its generalized approach of providing real-time traveler information from crowdsourced data