Heart-rate-measurement-using-camera

Abstract

• Heart Rate (HR) is one of the most important Physiological parameter and a vital indicator of people‘s physiological state
• A non-contact based system to measure Heart Rate: real-time application using camera
• Principal: extract heart rate information from facial skin color variation caused by blood circulation
• Application: monitoring drivers‘ physiological state

Methods

• Detect face, align and get ROI using facial landmarks
• Apply band pass filter with fl = 0.8 Hz and fh = 3 Hz, which are 48 and 180 bpm respectively
• Average color value of ROI in each frame is calculate pushed to a data buffer which is 150 in length
• FFT the data buffer. The highest peak is Heart rate
• Amplify color to make the color variation visible

Requirements

pip install -r requirements.txt

Implementation

python GUI.py

• In case of plotting graphs, run "graph_plot.py"
• For the Eulerian Video Magnification implementation, run "amplify_color.py"

Results

• Data from a specialized device, Compact 5 medical Econet, is used for the ground truth. In certain circumstances, the Heart rate values measured using the application and the device are the same

Reference

• Real Time Heart Rate Monitoring From Facial RGB Color Video Using Webcam by H. Rahman, M.U. Ahmed, S. Begum, P. Funk
• Remote Monitoring of Heart Rate using Multispectral Imaging in Group 2, 18-551, Spring 2015 by Michael Kellman Carnegie (Mellon University), Sophia Zikanova (Carnegie Mellon University) and Bryan Phipps (Carnegie Mellon University)
• Non-contact, automated cardiac pulse measurements using video imaging and blind source separation by Ming-Zher Poh, Daniel J. McDuff, and Rosalind W. Picard
• Camera-based Heart Rate Monitoring by Janus Nørtoft Jensen and Morten Hannemose
• Graphs plotting is based on https://github.com/thearn/webcam-pulse-detector
• https://www.pyimagesearch.com/2017/04/03/facial-landmarks-dlib-opencv-python/

Note

• Application can only detect HR for 1 people at a time
• Sudden change can cause incorrect HR calculation. In the most case, HR can be correctly detected after 10 seconds being stable infront of the camera
• This github project is for study purpose only. For other purposes, please contact me at [email protected]