Please use this identifier to cite or link to this item:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/84548
Title: | 基於視覺之睡姿分類及其應用 Vision-Based Sleep Posture Classification and Its Applications |
Authors: | Yan-Ying Li 李彥瑩 |
Advisor: | 洪一平(Yi-Ping Hung) 洪一平(Yi-Ping Hung | hung@csie.ntu.edu.tw | ), |
Keyword: | 睡姿分類,睡眠監控,影像摘要, sleep posture classification,sleep monitoring,video summarization, |
Publication Year : | 2022 |
Degree: | 博士 |
Abstract: | 睡眠障礙是現代社會許多人普遍困擾的問題,對於多數人的生活更是影響甚鉅。許多研究顯示憂鬱症、心臟病、身體疼痛和記憶力衰退是罹患失眠症者的普遍症狀,而肥胖症、關節炎、糖尿病和骨質疏鬆等常見病症更是和睡眠問題息息相關。研究還指出睡眠姿勢和睡眠品質高度相關。維持良好的睡眠姿勢有助於提升好的睡眠品質,而好的睡眠品質更是維持健康生活不可或缺的要件。因此,如何運用現代科技來記錄和監控睡眠姿勢和狀態,改善睡眠品質是健康領域的一項重要研究課題。 本研究著重在睡姿分類演算法以及睡眠監控和睡眠影像摘要系統的應用和開發。針對睡姿分類演算法,我們採用非侵入式的攝影機來記錄睡眠資訊,分辨睡眠姿勢。我們分別針對深度影像和非深度影像做睡姿分類演算法研究。考量設備的成本和普及性,我們使用一般家用監視器來開發睡眠監控系統。該設備不會對睡眠者造成干擾,可自動分析使用者的睡姿。 此外,有鑑於睡眠紀錄影像過於冗長,使用者需要耗費長時間和精力來瀏覽,我們開發了自動睡眠摘要系統,使用者可在短時間內瀏覽睡眠摘要。透過睡眠摘要及睡姿轉換分佈圖,使用者可了解自己睡眠中的狀態,並進一步調整睡眠習慣,增進睡眠品質。 最後,我們對開發的演算法和系統進行驗證,並跟其他研究者的方法做比較,皆獲得較高的準確度和較好的結果。驗證結果顯示,我們提出的系統和方法不僅能通過實驗室的各種情境測試,也適用於居家的睡眠情境。 Sleep quality is known to have a considerable impact on human health. Recent research shows that head and body pose are vital in affecting sleep quality. This dissertation aims to propose the sleep posture classification methods using deep learning and develop a sleep monitoring and summarization system. For sleep posture classification, the methods are studied with and without depth images. Furthermore, a deep multi-task learning network is proposed to perform head and upper-body detection and pose classification during sleep. For sleep monitoring and summarization, a system is developed to simultaneously detect and classify upper-body pose and head pose during sleep. It is a contact-free home security camera-based monitoring system that can work on remote subjects, as it uses images captured by a home security camera. In addition, a fully automatic video summarization method tailored to sleep monitoring is introduced. In addition to recognizing sleep posture, synopses of sleep video with keyframe and video view are provided. This information can help users better understand their sleep habits and find ways to improve and maintain sleep health. Experimental results show that our multi-task model achieves an average of 92.5% accuracy on challenging datasets, yields the best performance compared to the other methods, and obtains 91.7% accuracy on the real-life overnight sleep data. The proposed system can be applied reliably to extensive public sleep data with various covering conditions and is robust to real-life overnight sleep data. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/84548 |
DOI: | 10.6342/NTU202203514 |
Fulltext Rights: | 同意授權(限校園內公開) |
metadata.dc.date.embargo-lift: | 2022-09-26 |
Appears in Collections: | 資訊工程學系 |
Files in This Item:
File | Size | Format | |
---|---|---|---|
U0001-1709202221013900.pdf Access limited in NTU ip range | 5.79 MB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.