Eco-friendly Naturalistic Vehicular Sensing and Driving Behaviour Profiling

Internet of Things (IoT) technologies are spurring of serious games that support training directly in the field. This PhD implements field user performance evaluators usable in reality-enhanced serious games (RESGs) for promoting fuel-efficient driving. This work proposes two modules – that have been implemented by processing information related to fuel-efficient driving – to be employed as real-time virtual sensors in RESGS. The first module estimates and assesses instantly fuel consumption, where I compared the performance of three configured machine learning algorithms, support vector regression, random forest and artificial neural networks. The experiments show that the algorithms have similar performance and random forest slightly outperforms the others. The second module provides instant recommendations using fuzzy logic when inefficient driving patterns are detected. For the game design, I resorted to the on-board diagnostics II standard interface to diagnostic circulating information on vehicular buses for a wide diffusion of a game, avoiding sticking to manufacturer proprietary solutions. The approach has been implemented and tested with data from the enviroCar server site. The data is not calibrated for a specific car model and is recorded in different driving environments, which made the work challenging and robust for real-world conditions. The proposed approach to virtual sensor design is general and thus applicable to various application domains other than fuel-efficient driving. An important word of caution concerns users’ privacy, as the modules rely on sensitive data, and provide information that by no means should be misused.