AI-Based Driver Impairment Detection and Smart Vehicle Safety System

P Adhikya1, A Jahnavi2, V Sanjana3, S Venkata Datta Chandan4, Mrs L Lavanya5
1 Department of Computer Science and Engineering(AI&ML), Student, Sri Venkateswara College of Engineering
2Department of Computer Science and Engineering(AI&ML), Student, Sri Venkateswara College of Engineering
3Department of Computer Science and Engineering(AI&ML), Student, Sri Venkateswara College of Engineering
4Department of Computer Science and Engineering(AI&ML), Student, Sri Venkateswara College of Engineering
5Department of Computer Science and Engineering(AI&ML), Assistant Professor, Sri Venkateswara College ofEngineering

Abstract—Driver impairment arising from drowsiness, emotional distress, and sustained inattentiveness remains a leading contributor to road traffic fatalities worldwide. This paper presents SafeDrive AI, a real-time, vision-only driver monitoring framework employing computer vision and deep learning to assess driver alertness continuously during vehicle operation. A cabin-mounted camera captures live facial data processed using OpenCV to extract key behavioral indicators including eyelid closure frequency, head-pose variation, and facial expression dynamics. The extracted features are analyzed by a mini-Xception Convolutional Neural Network (CNN) trained on the FER-2013 facial expression dataset, classifying the driver’s affective state across seven emotion categories. A rule-based decision engine maps the CNN output and eye-state cues to one of four driver-status labels—SAFE, IMPAIRED (DROWSY), IMPAIRED (EMOTIONAL), or CRITICAL—triggering corresponding dashboard alerts. The framework operates exclusively on visual input via a Streamlit web application, eliminating dependency on physiological sensors or vehicle hardware modifications, ensuring cost-effective and scalable deployment across heterogeneous vehicle fleets.Keywords: Driver Impairment Detection; Computer Vision; OpenCV; Haar Cascade Classifier; FER-2013; Emotion Recognition; Mini-Xception CNN; Rule-Based Decision Engine; Streamlit.