Projects

Activity Classifier


A technology for classifying mobile users’ behaviors

 

Activity Classifier_ページ用

 

Activity Classifier is designed to tell what mobile phone users are currently doing—and where. It functions by mobilizing information about their vibration and action, which is collected by accelerometers and gyrosensors. Accelerometers are incorporated into many devices--to detect the direction of gravity, or to protect a hard disk drive in the event of a fall. Gyrosensors are in lots of devices as well, for such purposes as image stabilizer functionality in cameras. Notably, these sensors can tell what the user of the device is doing by recognizing component/frequency analysis of vibrations generated by such activities as walking, running, sitting down, getting on a train, or taking an elevator.

 



Devices like smartphones are normally kept in pockets and bags.

 

Some users carry their smartphones in their hands or hang them around their necks. Any of these approaches is fine. Activity Classifier does not quarrel with the device’s position or direction. Activity Classifier has the unique ability to deduce where the device is located – say, in a pocket or a bag -- by analyzing the sensor waveforms during walking for each device position.

Activity Classifier stemmed from joint research by Brian Clarkson of Sony CSL, and Masatomo Kurata and Makoto Murata, both of Sony Corp. After first proposed, the technology continued to be upgraded, and went into practical use in 2014 as the Lifelog application for SmartBand and Xperia. All of these accomplishments will eventually broaden Activity Classifier’s applications. For example, users will be provided automatically with information that matches their behavior and surroundings. They also will be able to share their actions through blogs or avatars, or be able to regularly monitor athletic workouts. Expectations are high for Activity Classifier to develop further as a key technology in wearable equipment.