2024-03-29T01:19:20Zhttps://ipsj.ixsq.nii.ac.jp/ej/?action=repository_oaipmhoai:ipsj.ixsq.nii.ac.jp:002107702023-11-14T00:51:14Z06164:06165:06640:10580
Preliminary Investigation of Unsupervised Factory Activity Recognition with Wearable Sensors via Temporal Structure of Multiple MotifsPreliminary Investigation of Unsupervised Factory Activity Recognition with Wearable Sensors via Temporal Structure of Multiple Motifseng行動認識http://id.nii.ac.jp/1001/00210664/Conference Paperhttps://ipsj.ixsq.nii.ac.jp/ej/?action=repository_action_common_download&item_id=210770&item_no=1&attribute_id=1&file_no=1Copyright (c) 2020 by the Information Processing Society of JapanGraduate School of Information Science and Technology, Osaka UniversityGraduate School of Information Science and Technology, Osaka UniversityCorporate Manufacturing Engineering Center, Toshiba CorporationGraduate School of Information Science and Technology, Osaka UniversityQingxin, XiaKorpela, JosephNamioka, YasuoMaekawa, TakuyaIn this study, we propose a robust unsupervised learning technique that makes use of two types of sensor data motifs to track the starting time of each operation in every iteration of work periods. A period motif only occurs once in each work period that is used to roughly detect the duration of work periods. An action motif occurs many times in each work period, corresponding to some basic actions in the period. A temporal structure is then constructed based on the temporal distances among motifs in the first period, which is used to improve motif tracking in the following periods as well as roughly detect the location of outliers. We run particle filters to track the starting time of operations and select a best particle series based on the extracted motifs. We evaluate the proposed method using sensor data collected from workers in actual factories and achieved state-of-the-art performance. In this study, we propose a robust unsupervised learning technique that makes use of two types of sensor data motifs to track the starting time of each operation in every iteration of work periods. A period motif only occurs once in each work period that is used to roughly detect the duration of work periods. An action motif occurs many times in each work period, corresponding to some basic actions in the period. A temporal structure is then constructed based on the temporal distances among motifs in the first period, which is used to improve motif tracking in the following periods as well as roughly detect the location of outliers. We run particle filters to track the starting time of operations and select a best particle series based on the extracted motifs. We evaluate the proposed method using sensor data collected from workers in actual factories and achieved state-of-the-art performance.マルチメディア,分散協調とモバイルシンポジウム2076論文集20203743812020-06-172021-04-16