加速度センサを用いた人間の行動推定に関する研究

抄録

Recently, electric devices such as smart phone and wearable devices have been rapidly became widespread following the evolution of digital, technology, sensor performance and application software technology. Various sensors are mounted on those devices, including GPS, accelerometer, gyro sensor, light sensor, magnetic sensor etc. Monitoring data of daily life is easily obtained by carrying these devices. Personal usages of these devices can allow us to do an activity monitoring or a specific health issue monitoring, a fitness tracking and a gauging for alertness and energy levels. For these personal usages, it is important to classify accurately human activities using signals from sensors imbedded their electric devices and machine learning technology may be an effective technique. In the study, we carried out the activity estimation using the acceleration data obtained from smart phone. The human activities are classified into six patterns of running, walking, standing, sitting, and two types of climbing stairs. The smart phone for data collecting is attached outside of a thigh. As a result, classifier with best performance was the recognition system adopting machine learning technique of random forest and if several problems are solved, the system can improve the performance to the level causing no problem in practical use.