A Driver Detection Method by Means of Explainable Deep Learning

The introduction of electronics in modern vehicles has sparked the inventiveness of thieves, who are always finding new ways to steal cars. With the aim to avoid vehicle theft, in this paper we propose a method aimed to continuously detect the driver when the driving session is in progress i.e., by providing a silent and continuous way to authenticate the (authorized) driver to the vehicle (and to continue to authenticate him/her while driving). We analyse a set of features extracted from the vehicle controller area network that are considered as input for several deep learning networks, aimed to discriminate between different drivers. A real-world path in Korea performed by four different drivers is used in the experimental analysis, by showing promising results: as a matter of fact, the proposed method obtains a precision equal to 0.906 and a recall of 0.887 with the MobileNet model in driver detection.