Smart Watch

Movement Sensing

I used a Inertial Measurement Unit (IMU) to capture the movement of the device. The IMU has 9 signal outputs: gyroscope (3 axis), accelerometer (3 axis), and magnetometer (3 axis). Teensy is used to read the signal output from IMU (MPU-9250).

Data Filtering

Right window size and low-passed filter are chosen to process the 9 outputs data from IMU, so that the data can be further processed in the real time.

ML Model Training

Generated training dataset for ML model in Weka (a machine learning library based on Java). Tried different ML algorithms and optimized the parameters to achieve higher accuracy. I found out that KNN (K=1) can achieve the highest accuracy.

ML Model Application

Apply the model to achieve the real time movement detection.

Smart Watch Detection

Repeated the processes and achieved the movements (forward-backward, up-down, left-right, watch click, arm tap, rotate) detection of a smart watch.