Automatic Handwriting Inference via Motion Sensor Embedded Wrist-worn Device

The rise of sensor-equipped Internet of Things devices has flourished the development of smart sensing systems. Handwriting sensing as an important enabler for applications in smart homes and smart educations, etc, is a challenging problem due to the noise in the sensor data collected from common off-the-shelf MEMS equipped devices, as well as the diversities in the writing style of different users. In this paper, to encounter these problems, we propose a novel smartwatch based handwriting sensing system which is unobtrusive to use and achieves high recognition accuracy. To achieve a robust recognition accuracy, we incorporate a data pre-processing module and a novel feature fusion model which employs PCA for time domain and frequency domain feature fusion. A deep learning model is then used to achieve a high recognition performance with 92.42% accuracy on average. The results show that the system has potential to support future smart handwriting sensing applications.