面向老年照护的实时困倦检测
计算机视觉与模式识别
2020-10-22 v1 人工智能
摘要
本文的主要焦点是产生一个从视频中提取困倦信息以协助独居老年人的概念验证。为量化随时间变化的打哈欠、眼睑与头部运动,我们从采集视频中提取 3000 张图像,用于集成 OpenCV 库的深学习模型的训练与测试。眼睑与嘴张开/闭合状态的分类准确率达到 94.3%–97.2%。对带有生成 3D 坐标叠加的视频中头部运动的视觉检查,显示出所采集数据(偏航、翻滚与俯仰)中清晰的时空模式。该困倦信息作为时间序列的提取方法可应用于其他情境,包括支持先前在隐私保护增强型辅导、运动康复以及医疗大数据平台集成方面的工作。
引用
@article{arxiv.2010.10771,
title = {Towards Real-time Drowsiness Detection for Elderly Care},
author = {Boris Bačić and Jason Zhang},
journal= {arXiv preprint arXiv:2010.10771},
year = {2020}
}
备注
This unpublished paper was accepted by the Conference on Innovative Technologies in Intelligent Systems & Industrial Applications (CITISIA 2020) [https://ieee-citisia.org] and uploaded to ArXiv.org preprint server. The camera-ready copy with DOI should be available in IEEE Xplore sometimes after the conference presentation and copyright transfer to IEEE