Drowsiness Detection System in Drivers Using Micro-Maneuvers on the Steering Wheel and Machine Learning to Prevent Nighttime Traffic Accidents

Heredia C.A.; Huacasi F.J.A.; Apfata Y.E.V.; Suarez J.T. · SSRG International Journal of Electrical and Electronics Engineering (2024) · DOI: 10.14445/23488379/ijeee-v11i11p112

Resumen

Traditional techniques to identify drowsiness, such as physiological sensors and camera-based eye tracking, often encounter difficulties in practice due to their intrusiveness, cost and vulnerability to other influences. This paper presents a novel non-invasive sleep detection system that accurately detects driver sleep indicators by combining machine learning approaches with driving micro-maneuvers. This method uses high-precision turning data obtained from sensors embedded in the vehicle's steering wheel. The system uses this data to extract characteristics related to driver drowsiness, relating small variations in the turning of the steering wheel, which, depending on whether they are present or not, determine the driver's state and classify it as "alert"or "drowsy"thanks to the use of sophisticated autonomic learning techniques, such as Support Vector Machines (SVM) and Convolutional Neural Networks (CNN). In order to validate the system, several tests were performed in controlled scenarios, with various drivers with different states of drowsiness previously verified. The tests showed that the proposed system obtained an overall accuracy rate of more than 92%, which allows it to be used in real scenarios without discomfort to the driver as it is a non-invasive measurement technique.

Instituciones: Universidad Nacional de San Agustín de Arequipa