Abstract & Overview
Understanding driving behavior – even in the rapid emergence of automation - remains in the spotlight, for decomposing complex driving dynamics, enabling the development of user-friendly and acceptable autonomous vehicles and ensuring the safe co-existence of autonomous and conventional vehicles on the road. Mobile crowdsensing has emerged as a means to understand and model driving behavior. Although the advantages of collecting data through smartphones are many (speed, accuracy, low cost etc.), the challenges including, but do not limited to, the preparation rate, the processing needs, as well as the methodological, legislative and security issues, are significant. The present paper aims to review the research dedicated to analyzing driving behavior based on smartphone sensors’ data streams. We first establish an inclusive stepwise framework to describe the path from data collection to informed decision making. Next, the existing literature is thoroughly analyzed and challenges in relation to data collection and data mining practices are critically discussed placing particular emphasis on the limitations and concerns regarding the use of mobile phones for driving data collection, as well as using crowd sensed data for feature extraction. Subsequently, modeling driving behavior practices and end-to-end solutions for driver assistance and recommendation systems are also reviewed. The paper ends with a discussion on the most critical challenges arising from the literature and future research steps.