Human Activity Recognition Based on Accelerometer Vibrations Using Artificial Neural Network
Meo Vincent C. Caya, Analyn Niere Yumang, Jhayvee V. Arai, John Daryll A. Ninofranco, Kenneth Aaron S. Yap · 2019
In recent years, activity recognition gained popularity in the field of wearable applications. One field that benefitted from the development of activity recognition is medical health and monitoring. In the Philippines, FNRI stated that Non-Communicable Diseases (NCDs) caused 67% of the total death in the country. As such, companies started producing wearable devices that can measure burned calories based on the inputted exercise of the user which can help especially the health conscious. Although diet and exercise cannot be considered as cure, it can be an assistance and prevention. In this paper, the researchers aimed to recognize the activity of the user using a wrist worn device. The device is composed of a raspberry pi and accelerometer. In order to recognize the activity, the vibration signals from the accelerometer were analyse and processed based on neural network analysis which was used to predict the activity done by the user. Once the activity is predicted it will compute the corresponding burned calories per activity. The researchers conducted the experiment composed of 50 persons for training and 30 persons for testing of the prediction model wherein the subjects were asked to perform an exercise within 20 seconds equivalent to 1 trial. The subjects performed each activity good for 60 seconds which is equivalent to 3 trials for each activity, 12 trials in total for 1 person. The test resulted with a 91.11% accuracy based on a 360-trial experiment.