Using Computer Vision for Automated Fall Risk Assessment
Ng, Becky 1, 2, ; Mihailidis, Alex 1,2 ;
1. University of Toronto; 2. Toronto Rehabilitation Intsitute
Recent advances in machine learning, such as the development of deep learning algorithms, have resulted in great advances in the effectiveness of machine learning applications. One such application is called human pose estimation, which is a computer’s ability to take estimate the pixel locations of a person’s joints and body parts when given an input image or video. Having a computer understand how a person is moving is a novel concept with many implications. Human pose estimation has the potential to automate some clinical procedures, especially in the field of rehabilitation. One such procedure is the fall risk assessment of elderly dementia patients. While the current method requires a physiotherapist to perform the assessments, a proposed device will be able to ambiently record video of these patients in their everyday movements and use human pose estimation to complete this assessment automatically.