Recognizing User’s Activity based on Analyzing Low-Power Wireless Radio with Machine Learning Techinque: Bluetooth Headphone as Example

• Main Authors: CHEN, GUAN-HENG, 陳冠亨
• Other Authors: LEE, HUANG-CHEN
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/7nbmac

碩士 === 國立中正大學 === 通訊工程研究所 === 106 === Bluetooth technology frees the limitation of earphone wires, but suffered from wireless interference and attenuation from other radio sources and obstacles, results in link loss. For a master device (i.e., a smartphone), it cannot know the exact state of its slave device (i.e., a Bluetooth earphone) while the link loss occurred. In this situation, if the master can collect enough information to determine the slave’s state and knowing the users’ activities at the moment of link loss, then it may be appropriate respond to the link loss more quickly and correctly. This study focused on the usage of Bluetooth earphones and proposed a method to determine user’s activity based on the characteristics of wireless communication and earphone’s physical data, e.g. radio signal strength and acceleration. Machine learning techniques were applied to determine what caused link loss. Therefore, the smartphone may actively disconnect the Bluetooth link to save energy while the user leaves the earphone home intentionally. This study defined five scenarios of user’s activities to build classifier models, and the results show SVM classifier can determine user’s activities with 7.5-second data while the accuracy can achieve as high as 90%.