Real-time Action Recognition and Fall Detection Based on Smartphone
Yunkun Ning, Shiwei Hu, Xiaofen Nie, Shengyun Liang, Huiqi Li, Guoru Zhao · 2018
This paper presents a smartphone application which has realized action recognition and fall detection. The application identifies the holding pattern of smartphone by the data of light sensor, distance sensor and accelerometer sensor, which reduce the impact of recognition resulting from the smartphone's different positions. And then the application uses data collected from the acceleration sensor, the direction angle sensor and the gyro sensor to distinguish fall from daily actions. The results of human motion recognition are uploaded to the server. For the purpose of real time, the network stability of the application is improved by the method of multi-layer detection based on heartbeat packet. Experiments prove that the way of improving network stability can reduce the rate of losing packet. The accuracy of action recognition achieves more than 90%.