Human activities recognition using machine learning for elderly in smart living environment

Abstract

In past years, the application Human Activity Recognition also known as HAR has grown significantly in the smart living environment. In the past, researches and studies of sensor data acquisition for activity recognition was costly and difficult as they required custom hardware. Now with the improvement of technology, smart phones and other personal health tracking devices are affordable and widely used. Therefore, data collection from these sensors has become cheaper and hence more commonly seen, resulting in more studies on problem solving using HAR. However, despite of the recent advancement of HAR, a population group that is frequently neglected is the elderly. As the aging population increases, the risk of health problems also increases. In order to reduce those health problems, the elderly are encouraged to have a healthy and balanced lifestyle. In order to do that, we need to have the ability to understand and allocate the lifestyle most suitable for them. In this project, we will be looking at introducing yoga as a form of healthy activity for the elderly. The project will focus on the task of human activity recognition on basic yoga movements from accelerometer data. The training model weighted KNN has achieved 93.3% accuracy on 6 yoga poses: mountain; tree; triangle; bridge; butterfly; cat-cow pose. Other training algorithms such as SVM and NB have also been explored and studied in this report.