“My autonomous car is an elephant”: A Machine Learning based Detector for Implausible Dimension

Abstract

Connected and Automated Vehicle is the next goal for car manufacturers towards traffic safety and efficiency. To ensure safety, automotive applications rely on data acquired through vehicular communication and locally embedded sensors. Among these data, classification data permit the autonomous vehicle to decide to pass another vehicle according to not only its dynamic but also its length and width. Unlike sensors which are prone to measurement errors, vehicular communication allows others connected vehicles to provide their exact dimension values based on car manufacturer specification. However, this fact assumes that other road users may not lie. Currently, researchers focus on malicious mobility data but none focus on classification data within V2X message. Therefore, this paper proposes a misbehavior classifier related to classification data for multiple types of road users. Thus, we compare four methods that include a threshold classifier (MinMax) and three machine learning algorithms.