Motion Recognition by Using a Stacked Autoencoder-Based Deep Learning Algorithm with Smart Phones

摘要

Due to the increasing machine learning applications in mobile health and security monitoring scenarios, human motion recognition by sensor devices has received remarkable attention from both academic and engineering fields. In this paper, we propose a motion recognition scheme by using a stacked autoencoder based deep learning algorithm with smart phones. Since common sensors such as gravity sensors, accelerometers, gyroscopes, linear accelerometers and magnetometers have been already equipped in Android or iOS based smart phones, the sensor data can be easily recorded by the smart phone that an experimenter carries around. A stacked autoencoder based deep learning algorithm is employed here for data classification so as to precisely recognize several basic motions that are standing, walking, sitting, running, going upstairs and going downstairs, respectively. Experimental results indicate that the stacked autoencoder based deep learning algorithm achieves higher accuracy for human motion recognition than traditional neural network methods.