AFAR: A Real-Time Vision-based Activity Monitoring and Fall Detection Framework using 1D Convolutional Neural Networks
Jessie James P. Suarez, Nathaniel S. Orillaza, Prospero C. Naval
Abstract
In recent years, there has been an increased interest in the use of telemedicine as an option to avail proper healthcare. However, one of the main issues of activity monitoring and fall detection in telehealth systems is the scalability of the technology for areas with inadequate technology infrastructure. As a potential solution, this study proposes an efficient activity monitoring and fall detection framework which can run real-time on CPU devices. In comparison to previous works, this study makes use of an efficient pose estimator called MediaPipe and leverages the pose joints as the main inputs of the model for activity monitoring and fall detection. This allows the framework to be used on cost-effective devices. To ensure the quality of the framework, it was evaluated on three (3) publicly available datasets: Adhikari Dataset, UP Fall Dataset, and UR Fall Dataset by looking at accuracy, precision, recall, and F1 scores. Based from the results, the framework was able to achieve both state-of-the-art and real-time performance on these datasets.