Risky Driving Behavior Detection Using In-Vehicle WiFi Signals

Risky driving is a leading cause of traffic accidents all over the world. Major factors contributing to risky driving include driver distraction (e.g. smart-phone usage) and an onset of driver drowsiness. As a result, improving road-safety requires modeling and detecting in-vehicle conditions indicative of risky driver behavior. Previous works towards driver behavior modeling and classification possess limitations such as increased expenditure and/or additional external infrastructure requirements. In this paper, we propose a novel driver behavior detection framework that at its core makes use of in-vehicle Wi-Fi signals. The main premise of the proposed framework is that risky behavior induces specific differentiable patterns in received WiFi signals - patterns that are then exploited to build an efficient classification system. More specifically, we classify drowsy and inattentive driving into four main gestures that reflect unsafe driving. These gestures include (a) Yawning, (b) Head Jerks, (c) Sideways motion, and (d) Smart-phone usage. Channel State Information (CSI) is used to determine the driver's gesture which is further used for behavior modeling. A number of low computational features are used to classify among different gestures. Experiments are performed in a real in-vehicle environment using software-defined radios. Using data collected from these experiments, a simple ensemble classifier is implemented which is shown to achieve an average accuracy of 88.64%.