Automated Fall Detection via Webcam

Research Hypothesis: Video from a webcam may be processed using current ml5.ps posenet library to capture skeleton of a human. An algorithm may be applied to this skeleton to achieve real-time detection of a fall event. Motivation: Numerous fall related injuries occur every day. In the event the person falling is alone, an automated detection system could be used to alert others in a timely manner.

Novelty: This uses a single webcam. It uses existing ml5.js and p5.js libraries. The algorithms being investigated are derived from papers using accelerometers and other physically attached sensors. Additional data from the skeletal tracking is used to supplement the algorithms. Thus no extra sensors are required. Final implementation is targetted for a standalone hardware device composed of a camera, a computational-unit, and some additional components for connecting to various safety/security networks.

Results: Current results are limited. More advanced algorithms are being tested. This is an early proof of concept.

Live Demo: This requires an https connection. My SSL certificate sometimes expires before I can renew it. So you may encounter some browser questions... In all cases you will have to allow the browser access to your webcam. All processing is done locally. Nothing will be stored on this server. The record button is a toggle. Press it to begin recording. Press it again to stop. The recording is done using your computer's memory (RAM). So do not record for long durations. When you stop - it may ask you to 'download' the .webm file, it is really just writing it out from memory. demo