Computer Vision for Fatigue Detection on Vehicle Drive

This research evaluates four computer vision methodologies CNN, MediaPipe, Dlib, and YOLO for detecting driver drowsiness using a public dataset, according to a uniform preprocessing pipeline, dataset partitioning, and evaluation protocol. Frames undergo conversion to grayscale, histogram equalization, and resizing before to categorization. CNN attained the highest per-frame accuracy, succeeded by YOLO, MediaPipe, and Dlib; nevertheless, YOLO provided the optimal deployment balance, exhibiting near-peak accuracy, the swiftest inference speed, and the least accuracy reduction in simulated low-light environments. In summary, YOLO is optimally designed for real-time applications, whereas CNN is the most precise classifier.