Activity Recognition using Accelerometers
碩士 === 國立臺灣科技大學 === 工業管理系 === 105 === Accelerometer has been widely used as a tool for studying posture stability because of its convenience and inexpensive. Acceleration data can reflect the intensity of human movement and it’s a simple way to verify the accuracy of accelerometers by detecting the...
Main Authors: | , |
---|---|
Other Authors: | |
Format: | Others |
Language: | zh-TW |
Published: |
2017
|
Online Access: | http://ndltd.ncl.edu.tw/handle/v7bgmp |
id |
ndltd-TW-105NTUS5041087 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-TW-105NTUS50410872019-05-15T23:46:35Z http://ndltd.ncl.edu.tw/handle/v7bgmp Activity Recognition using Accelerometers 應用加速規於動作辨識之研究 Yu-Chia Chen 陳裕佳 碩士 國立臺灣科技大學 工業管理系 105 Accelerometer has been widely used as a tool for studying posture stability because of its convenience and inexpensive. Acceleration data can reflect the intensity of human movement and it’s a simple way to verify the accuracy of accelerometers by detecting the number of steps because the speed of each activity is significantly different. However, if you want to further study the quality of the movement, gait assessment cannot achieve this purpose. Besides, there will be different effects for accuracy by placing in different parts of the body. In this research, we used the feature extraction and multiscale entropy analysis to get the better performance of three different accelerometers (Arduino, Cavy, Curo) and then compare the classification of different positions (knee, ankle, back, wrist). Twenty participants performed three different activities (walk, jog, run) on the treadmill and ten kinds of features (including MSE) were calculated by Matlab, then we compare the classification by random forest classification method. The results showed that average, maximum, minimum, energy, and root mean square were the most appropriate features to analyze the acceleration data. We found that the accuracy would increase significantly after adding the MSE for one of the features. For comparing three different accelerometers, the accuracy of Arduino accelerometer was the highest and the multiscale entropy analysis showed that Arduino and Curo accelerometers could clearly distinguish the difference between the three activities. This study also found that the back and wrist were the most accurate locations and the MSE could be the features to classify the data as different activities. Bernard-C. Jiang 江行全 2017 學位論文 ; thesis 95 zh-TW |
collection |
NDLTD |
language |
zh-TW |
format |
Others
|
sources |
NDLTD |
description |
碩士 === 國立臺灣科技大學 === 工業管理系 === 105 === Accelerometer has been widely used as a tool for studying posture stability because of its convenience and inexpensive. Acceleration data can reflect the intensity of human movement and it’s a simple way to verify the accuracy of accelerometers by detecting the number of steps because the speed of each activity is significantly different. However, if you want to further study the quality of the movement, gait assessment cannot achieve this purpose. Besides, there will be different effects for accuracy by placing in different parts of the body.
In this research, we used the feature extraction and multiscale entropy analysis to get the better performance of three different accelerometers (Arduino, Cavy, Curo) and then compare the classification of different positions (knee, ankle, back, wrist). Twenty participants performed three different activities (walk, jog, run) on the treadmill and ten kinds of features (including MSE) were calculated by Matlab, then we compare the classification by random forest classification method. The results showed that average, maximum, minimum, energy, and root mean square were the most appropriate features to analyze the acceleration data. We found that the accuracy would increase significantly after adding the MSE for one of the features. For comparing three different accelerometers, the accuracy of Arduino accelerometer was the highest and the multiscale entropy analysis showed that Arduino and Curo accelerometers could clearly distinguish the difference between the three activities. This study also found that the back and wrist were the most accurate locations and the MSE could be the features to classify the data as different activities.
|
author2 |
Bernard-C. Jiang |
author_facet |
Bernard-C. Jiang Yu-Chia Chen 陳裕佳 |
author |
Yu-Chia Chen 陳裕佳 |
spellingShingle |
Yu-Chia Chen 陳裕佳 Activity Recognition using Accelerometers |
author_sort |
Yu-Chia Chen |
title |
Activity Recognition using Accelerometers |
title_short |
Activity Recognition using Accelerometers |
title_full |
Activity Recognition using Accelerometers |
title_fullStr |
Activity Recognition using Accelerometers |
title_full_unstemmed |
Activity Recognition using Accelerometers |
title_sort |
activity recognition using accelerometers |
publishDate |
2017 |
url |
http://ndltd.ncl.edu.tw/handle/v7bgmp |
work_keys_str_mv |
AT yuchiachen activityrecognitionusingaccelerometers AT chényùjiā activityrecognitionusingaccelerometers AT yuchiachen yīngyòngjiāsùguīyúdòngzuòbiànshízhīyánjiū AT chényùjiā yīngyòngjiāsùguīyúdòngzuòbiànshízhīyánjiū |
_version_ |
1719152844776407040 |