Driving Behaviour Detection Using Smart Steering Wheel: Supervised and Unsupervised Classification

Driving behaviours are the root cause of millions of road accidents every year. Aggressive and distracted driving are the two most important examples of driving misbehaviours. This paper proposes a real-time system to distinguish aggressive and distracted driving from safe driving in a five-second time window. The system monitors the driver's heart rate, hand position and force on the steering wheel, and the vehicle's inputs (steering wheel angle and gas/brake pedal position). Data were collected from three different driving scenarios by recruiting five participants using a driving simulator. The performance of various supervised classification methods including KNN, SVM, Neural Network, and Decision Tree was compared in which Gaussian SVM featured the highest accuracy at 95.4%. To address labelling challenges for supervised learning, we assessed the performance of unsupervised clustering methods, K-mean and Fuzzy C-mean (FCM), in the classification of collected data, where the FCM exhibited an accuracy of 81.16% in separating distracted and aggressive driving from safe driving. The proposed sensing and clustering method provides a suitable real-time assessment of the quality of driving while the fuzzy outputs can be used in designing different assertion levels and taking proper actions in assisted driving systems to minimize the risk of accidents.