MSE Capstone Project
In my final year of undergraduate engineering studies, I completed an 8-month capstone project. This project was completed with a multidisciplinary team of engineering, health science, and business and marketing students. With the help of SFU's Charles Chang Institute for Entrepreneurship, we were able to develop a prototype of a fall detection system. The system uses computer vision to detect falls, and sends an alert to a caregiver. Our team MAD Technology 2nd Place at the 2022 SFU OppFest in the Health & Wellness category.
Unlike most other projects I completed in school, this project was unique because we approached it as a technology and business venture. Before the technical aspects of the project, we needed to find a problem to solve. I conducted market research, and interviewed doctors, seniors, and caregivers to find a problem that was worth solving. After many interviews, we found that falls were a major problem for seniors. We then conducted a literature review to find the best way to detect falls. Current solutions on the market use sensors such as Inertial Measurement Units (IMUs) and pressure sensors. These sensors are expensive and require a lot of maintenance. We decided to use computer vision to detect falls because it is cheaper, more accurate, and requires less maintenance once built. It was a novel approach to the problem, and our team's vision was supported by the numerous surveys we conducted where senior users complained about their discomfort, forgetfulness and overall unwillingness to wear pendants or smart watches at their age. We were able to develop a prototype that could detect falls with 90% accuracy. We pitched this project to a small panel of investors and secured $5000 in funding to move our project forward.
Our project focused on senior users who rely on walkers to move from A to B. We proposed our accessory which sits on top of walkers and monitors the senior via camera. A small User Interface and emergency protocols resulted in a promising product. We developed and trained our own algorithm using video datasets found online. We built a physical working prototype that could detect falls with 90% accuracy.