Toward self-policing: Detecting drunk driving behaviors through sampling CAN bus data

In this paper, we propose a method to detect drunk driving patterns using only basic car sensors, available through off the shelf OBD-II dongles. The sensor data include standard OnBoard Diagnostic sensor information along with an accelerometer sensor and GPS coordinates which are provided by the dongle. We collect the information through drive tests of normal driving behavior and controlled drunk driving behavior. The controlled driving emulation reflects the most common cues relating to drunk driving. After datasets are collected, a window-based approached was used for data smoothing and feature extraction. Finally, our approach makes use of a machine learning algorithm (Logistic Regression) for classification to achieve an accuracy of 82%.