Please use this identifier to cite or link to this item: http://rportal.lib.ntnu.edu.tw:80/handle/20.500.12235/111688
Title: 以智慧椅墊進行坐姿分析之研究
Sitting Postures Analysis Using Smart Cushion
Authors: 李忠謀
Greg Lee
張雅婷
Chang, Ya-Ting
Keywords: 物聯網
壓力感測器
機器學習
深度學習
Arduino
Internet of Things
pressure sensor
machine learning
deep learning
Issue Date: 2019
Abstract: 在現代社會中,大部分人的生活型態,不論是工作或者休息,往往有很長的時間維 持坐姿。近年來有越來越多的疾病被證實與久坐有關。許多人認為坐姿是種休息的姿 勢,但研究中指出,比起站姿與躺姿,坐姿讓椎間盤承受的壓力更大,而不適當的坐 姿則更提升了椎間盤的壓力。 由於久坐逐漸成為現代人的生活習慣,所以適當的坐姿就顯得格外的重要。在日常 生活中,不適當的坐姿對於大多數的人而言,屬於較為舒適的姿勢,所以往往無心注 意自己的坐姿是否適當。故須透過工具協助來了解自己的坐姿情況。本研究設計一智 慧椅墊之雛形,旨在透過較低的成本 Arduino開發版與少量的壓力感測器,並且準確 的分類使用者的坐姿。 過去使用壓力感測器進行坐姿分類的相關研究中,透過傳統的機器學習方法進行坐 姿的分類,且使用較多數量的感器收集各類坐姿的資料。準確率落在百分之八十至百 分之九十。本研究使用一種傳統機器學習演算法與兩種深度學習之方法進行實驗,找 出適合進行坐姿分類之方式,並以特徵選擇實驗找到能夠準確分類坐姿之感測器數量 及擺放方式。 本研究除了使用限制坐姿使用資料進行坐姿分類模型訓練以及評估初步的分類結 果,並透過實際座椅使用情況資料,再次檢視此智慧椅墊在實際使用情形下,亦能有 良好的做姿分類表現。透過智慧椅墊設計實驗與特徵選擇實驗,本研究完成一智慧椅 墊,使用少量的感測器與基礎的物聯網開發板,降低了硬體成本,達成良好的坐姿分 類表現。
Most people's lifestyles in modern society, no matter they are working or resting, often maintain a prolong time as “sitting” posture. During the past years, more and more diseases have been confirmed to be related to sedentary. Many people regard sitting posture as a relaxing posture, but the research pointed out that sitting posture put much more pressure on the intervertebral disc than standing or lying posture, the improper sitting posture may increases the pressure of the intervertebral disc as well. Since sedentary sitting has gradually become a habit of most of people, proper sitting posture become more and more important. Improper sitting posture has been regard as a more comfortable posture for most of people, and it is often hard to pay attention to whether or not our sitting posture is appropriate in daily life. Therefore, we may need a tool help us to understand our sitting posture. The aim of this study is to design a prototype of a smart cushion which can classify the user's sitting position accurately, through Arduino101 and a small number of pressure sensors. Compare with the similar related work, some studies use the traditional machine learning method to classify sitting postures, also a large number of sensors are used to collect data of different sitting postures. The accuracy rate falls between 80% and 90%. In our study, we demonstrate one traditional machine learning algorithm and two methods of deep learning in order to find out the suitable method for sitting posture classification, through the feature selection experiment we can make sure the smallest number of sensors to classify the sitting posture accurately. In addition to using the limited sitting posture data to train the classification model and evaluate the preliminary classification results, our study also examines the smart cushion under the actual using situation data, and reach a favorable classification performance. Through the smart cushion design experiment and feature selection experiment, this study propose a smart cushion, using small number of sensors and the basic IoT development board, reducing the hardware cost and achieving a good sitting posture classification performance.
URI: http://etds.lib.ntnu.edu.tw/cgi-bin/gs32/gsweb.cgi?o=dstdcdr&s=id=%22G060547019S%22.&
http://rportal.lib.ntnu.edu.tw:80/handle/20.500.12235/111688
Other Identifiers: G060547019S
Appears in Collections:學位論文

Files in This Item:
File SizeFormat 
060547019s01.pdf3.43 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.