UC Berkeley

Technologies for Mobile ITS Applications and Safer Driving

by

Christian Georges Manasseh

Doctor of Philosophy in Engineering - Civil and Environmental Engineering

University of California, Berkeley

Professor Raja Sengupta, Chair

This thesis presents a user application for safer driving. The technology to enable such an application is discussed and evaluated for the more general use in ITS mobile applications. Two main problems present themselves when building user-centric ITS mobile applications: the first has to do with knowing the state of the environment variables as captured by relevant sensing hardware, and the second has to do with relaying the right amount of information to the user to achieve the purpose of the application. For the first, we present a middleware architecture that abstracts sensor data into an HTTP interface to allow data consumers to discover and bind to data producers. For the second we present user preference adaptive services that would enable the application to learn from past user experience and adapt its user interaction to meet user preferences. Both research efforts are combined to field test a Smartphone traffic safety application system that is enabled by the middleware and that leverages the user preference adaptive services to enhance user experience and improve driver safety.

The middleware is implemented and tested on three types of traffic mobility and safety applications to measure its performance. The performance measurements show the middleware to be efficient enough for road safety and congestion relief applications by limiting the overhead on the system to the order of 100 msec.

A field test consisting of 9 drivers in the San Francisco Bay Area is conducted using a Smartphone traffic safety application built on the middleware. Results from the field test show that driver safety on the highway can be improved through a soft-safety warning of slow traffic 1-mile ahead on the driver's route. Data from the field test is then fed to a couple of user preference adaptive services to improve user experience of the Smartphone application. In the first service we construct a Support Vector Machine learning engine for each driver and predict to 75% accuracy their classification of favorable vs. non-favorable alerts. In the second service, we construct a Decision Tree model for the driver's previous destinations and predict with 96% accuracy their destination given their current location on the road, time of day, and day of week.