Application of CNN for Human Activity Recognition with FFT Spectrogram of Acceleration and Gyro Sensors

At the SHL recognition challenge 2018, Team Tesaguri developed a human activity recognition method. First, we obtained the FFT spectrogram from 60-second acceleration and gyro sensor data for each of six axes. A five-second sliding window was used for FFT processing. About 70% of the spectrogram figures from the Sussex-Huawei Locomotion-Transportation dataset were used for training data. Our model was based on CNN using FFT spectrogram images. After training for 50 epochs, F-measure was about 90% for acceleration data and 85% for gyro data. Next, considering the results of each sensor axis, to improve the recognition rate, we combined the information of multiple sensors. Specifically, we synthesized new images by combining the FFT spectrogram figures of two axes and the best combination condition was examined by correlation analysis. The highest score, 93% recognition, came from the vertically arranged images derived from the norm of acceleration and the y-axis gyro.