Open-Set Driver Identification System Based on Metric Learning with Driving Situation Awareness

Open-set driver identification system uses a metric learning-based neural network to identify drivers through similarity comparison between drivers' driving styles. The neural network embeds the driving style similarity into a vector, even for drivers not seen during training. Thus, the open-set system can freely add drivers to be identified in a quick and easy way without retraining the neural network. However, a driver's driving style is subject to change depending on various driving situations. Due to these changes in driving style, driver identification via similarity comparison suffers performance degradation. In this paper, we propose a novel open-set driver identification system with driving situation awareness. The proposed system is characterized by comparing the similarity of driving styles for each driving situation by utilizing driving situation-specialized neural networks. When enrolling a new driver, our system creates multiple IDs for each driving situation and identifies the driver with these driving situation -aware IDs. Our experiments on a naturalistic driving dataset show that our system with situation-aware IDs achieved superior accuracy compared to existing systems.