Human Activity Recognition Using Convolutional Neural Networks

Using smartphone sensors to recognize human activity may be advantageous due to the abundant volume of data that can be obtained. In this paper, we propose a sensor data based deep learning approach for recognizing human activity. Our proposed recognition method uses linear accelerometer (LAcc), gyroscope (Gyr), and magnetometer (Mag) sensors to perceive eight transportation and locomotion activities. The eight activities include: Still, Walk, Run, Bike, Bus, Car, Train, and Subway. In this study, the Sussex-Huawei Locomotion (SHL) Dataset of three participants are used to recognize the physical activities of the users. Fast Fourier Transform (FFT) spectrograms generated from the three axes of the LAcc, Gyr, and Mag sensor data are used as input data for our proposed Convolutional Neural Network (CNN) model. Experimental results on the task of human activity recognition demonstrated the effectiveness of our proposed user-independent approach over that of competitive baselines.