Details

Autor(en) / Beteiligte

• Lyu, Jie
• Zhang, Hui
• Yuan, Zejian

Titel

Joint Shape and Local Appearance Features for Real-Time Driver Drowsiness Detection

Ist Teil von

• Computer Vision – ACCV 2016 Workshops, p.178-194

Ort / Verlag

Cham: Springer International Publishing

Link zum Volltext

Quelle

Alma/SFX Local Collection

Beschreibungen/Notizen

• In this paper, we propose a framework to detect driver drowsiness from video sequences for an advanced driver assistance system. Our method extracts the effective facial descriptors to describe the drowsiness based on face alignment, and classifies the driver facial states via random forest (RF), finally short-term voting and long-term correlation are applied to output smooth results with long-term memory. In particular, the proposed descriptors can encode both shape and local appearance by the located facial landmarks, and utilize the information from multiple frames to enhance the reliability. The classification and alignment based on RF structure are very efficient for drowsiness detection. Our system can obtain 94% accuracy on our F-DDD dataset and 88.18% accuracy on the evaluating set of NTHU-DDD dataset, meanwhile, the implementation achieves 22 FPS for 640 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\times \!$$\end{document} 480 videos.