Project Overview

Non-invasive medical assessment and treatment has become increasingly in demand. In the context of healthcare, associated benefits include improved procedural efficiency, operational cost reduction, and more hygienic treatments for patients. However, many commercially available technologies for non-invasive monitoring of vital signs require direct physical contact with an apparatus. Existing camera-based solutions often require high-quality video footage of a single subject who remains still under ideal lighting conditions. The goal of H.A.M. cam is to create a robust multi-subject video-based heart activity monitor, that could be used in environments such as waiting rooms to improve the healthcare experience. 

Approach

We first considered and outlined a set of design constraints as well as functional and product specifications pertaining to our project scope of a video-based heart activity monitoring system to be used by medical personnel in hospital waiting rooms. 

To realize our idea, we investigated methods of facial tracking and recognition to enable multi-subject monitoring. For detecting heart rate and possible abnormal fluctuations, we explored different methods of intelligent sampling and integrated the state-of-the-art Eulerian Video Magnification framework into our solution. This framework made it possible to magnify subtle changes in skin-color, imperceptible by the naked human eye, to extract a person's pulse from video footage. We also experimented with lighting compensation techniques as a means to achieve greater robustness. The system's performance was validated against the measurements from a third-party heart rate sensor. The design of the solution evolved over time in an iterative process. Lastly, medium-fidelity prototypes for a graphical user interface were designed, providing a vision for a graphical user interface that medical personnel could use in a waiting room context to monitor heart activity information of patients.

ResultS

The final prototype could process pre-recorded video clips with multiple human faces and could extract heart rate measurements. The system was able to achieve an average error rate of about 8% across the test videos, however with a range of errors from 0% to 26% (calculated by comparing the beats-per-minute read-out from our solution against that of a third-party sensor). For an individual test video, parameters of the system could be tuned to improve the results to within the target range, often with an error of zero.

Course

Completed as a 4th year design project in the Systems Design Engineering program at the University of Waterloo. Done in collaboration with Audrey Chung and Jan Kulinski. 

Themes

Non-Invasive Medical Assessment, Remote Heart-Rate Monitoring, Video-Monitoring 

Skills & Technologies

Computer Vision, UI/UX Design

My Role

I worked on implementing multi-subject tracking, performed the data validation, and led the investigation and design of the medium-fidelity graphical user interface. I also designed the poster displayed at the project symposium (pictured above). 

Report

A.Chung, J. Kulinski, J. Leong. 2014. HEART ACTIVITY MONITOR CAMERA: HEART RATE ACQUISITION FROM VIDEO