Intelli Drive+ Multi-Modal Fatigue Detection And Adaptive Alert System

Abstract

Driver fatigue is a leading cause of road accidents worldwide, contributing to an estimated 20% of all traffic fatalities. Existing single-modality detection systems, which primarily rely on visual cues like eye closure, suffer from poor robustness under real-world conditions (e.g., low lighting, occlusions) and high false-positive rates due to a lack of personalization. This paper presents IntelliDrive+, a novel multi-modal fatigue detection system that integrates visual, vocal, and vehicle behavioral data to achieve accurate, real-time driver state monitoring. The system employs a hybrid deep learning architecture combining Vision Transformers (ViTs) for global spatial feature extraction and Convolutional Neural Network-Long Short-Term Memory (CNN-LSTM) networks for temporal sequence modeling. A key contribution is an adaptive online learning module that personalizes detection thresholds to individual driver baselines, significantly reducing false alarms. Implemented on resource-constrained edge devices (NVIDIA Jetson Nano), the prototype achieves 94% classification accuracy, an average latency of 200 ms, and an 85% user-perceived alert relevance rate. This work establishes a robust, scalable framework for intelligent, human-centered fatigue monitoring, advancing both road safety and personalized well-being.