講演抄録/キーワード |
講演名 |
2020-11-27 10:20
Detection of human activity based on hybrid deep learning model using a low-resolution infrared array sensor. ○Muthukumar K A・Mondher Bouazizi・Tomoaki Ohtsuki(Keio Univ.) SeMI2020-39 |
抄録 |
(和) |
(まだ登録されていません) |
(英) |
Artificial Intelligence (AI) plays a significant role in the healthcare industry. Many applications have been developed using AI in healthcare. Among these, activity detection is one of the most important applications. Many AI-based activity detection systems use conventional machine learning methods to detect various activities. In a conventional machine learning model, activity features are manually extracted and detected which presents one the main drawbacks of this family of techniques.. This report proposes an activity detection approach based on a hybrid deep learning model using a low-resolution infrared array sensor placed on the ceiling. The hybrid deep learning model automatically learns the features and detect the activity. Upon training, the classification is performed faster than that using conventional machine learning models.. The data collected from the infrared array sensor is classified using a CNN (Convolutional Neural Network) where each frame is individually classified. The CNN’s output is passed to the LSTM (Long Short Term Memory) for sequential classification with a time window size equal to five frames. The classification accuracy reach 96.60% and 97.74% for the CNN and the CNN+LSTM models, respectively, respectively. |
キーワード |
(和) |
/ / / / / / / |
(英) |
AI healthcare / activity detection / hybdrid deep learing / / / / / |
文献情報 |
信学技報, vol. 120, no. 261, SeMI2020-39, pp. 99-104, 2020年11月. |
資料番号 |
SeMI2020-39 |
発行日 |
2020-11-19 (SeMI) |
ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
PDFダウンロード |
SeMI2020-39 |