Sensor Context Estimation Technology Combining Physical and Semantic layers

2018 Fiscal Year Final Research Report

• Project/Area Number: 26280041
• Research Category: Grant-in-Aid for Scientific Research (B)
• Allocation Type: Partial Multi-year Fund
• Section: 一般
• Research Field: Multimedia database
• Research Institution: Kyushu Institute of Technology
• Principal Investigator: Inoue Sozo 九州工業大学, 大学院生命体工学研究科, 准教授 (90346825)
• Co-Investigator(Kenkyū-buntansha): 西田 健 九州工業大学, 大学院工学研究院, 准教授 (30346861)
• Project Period (FY): 2014-04-01 – 2019-03-31
• Keywords: 行動認識 / センサ行動認識

Outline of Final Research Achievements

In this research, we researched the technology to efficiently recognize and estimate human activity and placed situation using sensor devices from two layers of physical and semantic viewpoints. As a result, we achieved many results such as action recognition methods for unknown action classes, future prediction methods related to actions, machine learning and feature quantity optimization considering data loss, utilization of crowdsourcing and gamification in data collection, medical facilities and nursing facilities, large-scale activity data collection by nurses and caregivers, activity recognition using LoraWAN sensor, activity recognition considering time lag in activity labels, and transfer learning considering differences between users and homes. We received 22 awards for excellent dissertation.

Free Research Field

ユビキタスコンピューティング、データ工学

Academic Significance and Societal Importance of the Research Achievements

行動認識は、人間の行動を対象とするため、様々な社会応用が考えられるが、画像認識や音声認識といった他の認識分野に比べてまだ普及していない。その理由は、センサが非常に単純なものであることに比べて行動そのものは人によってかなり異なることに起因する。本研究はその問題に潜む様々な側面に光を当てて、一つずつ解決を試みることで、多くの受賞を得ることができた。