紀要論文

複数のウェアラブルデバイスから得られる慣性データとコンテキストデータを用いた行動認識

Human Activity Recognition Utilizing Inertial and Contextual Data from Multiple Wearable Devices

jpn

24321192

https://doi.org/10.15002/00022728

法政大学大学院情報科学研究科

法政大学大学院紀要. 情報科学研究科編

15

1

6

2020-03-24

Version of Record

This research has been aimed at human activity recognition (HAR) using both inertial data and context data from multiple wearables. Ten participants are recruited to conduct twelve daily activities such as walking, stepping stairs, typing, and writing, during which each participant wears multiple devices that are placed different parts in its body. The inertial data is captured from accelerometers and gyroscopes in the seven devices. The context data is special information that represents the state and environment situation to the user who conducts the twelve activities. The context data used in this study includes the WiFi access points’ IDs the wearable devices connected to, the atmospheric pressure levels when moving, the total changes in inertial data, and the devices’ positions in a body. The key in this research is how to organize the context information in some knowledge rules, which are then applied to divide activities into groups and detect a concrete activity together with machine learning algorithms such as SVM and k-NN. Both the probability distribution and the knowledge-based threshold are adopted in using the context. As a result, the recognition accuracy is 33% higher than that when only using inertial data, and 17% higher than that when using context data as feature together with inertial data features.

Article

AA12746425