samproell / yarppg

Yet another implementation of remote photoplethysmography (rPPG) in Python

Yet another rPPG

* This is a work in progress, do not blindly trust the results *

yarppg is yet another implementation of remote photoplethysmography in Python. Remote photoplethysmography (rPPG) refers to the camera-based measurement of a (blood volume) pulse signal. It works by detecting small changes in skin color, originating from the pulsation of blood¹.

Some other open-source implementations of rPPG:

• heartbeat: Live heart rate measurements. Written in C++ with a number of customization options (https://github.com/prouast/heartbeat)
• Bob's implementation of different rPPG algorithms: Python implementation of different publications for offline evaluation. (https://www.idiap.ch/software/bob/docs/bob/bob.rppg.base/master/)

The current default implementation uses the bandpass-filtered green channel averaged across a custom ROI suggested by Li et al.² Other features of their algorithm are a work in progress.

Installation and usage

In order to run this rPPG implementation, you first have to clone the repository and make sure all requirements are installed. Afterwards you can run the yarppg subfolder as a python package using the -m flag:

git clone https://github.com/SamProell/yarppg.git
cd yarppg
pip install -r requirements.txt
python -m yarppg  # from yarppg top-level directory

Alternatively, install the package using pip (from the top-level directory). After installing, you can simply call the provided entry point run-yarppg.

pip install -e .
run-yarppg

Command line options

There are a number of options available, when running yarppg:

Option	Default	Description
--detector	facemesh	ROI (face) detector - choices: {facemesh,caffe-dnn,haar,full}
--processor	LiCvpr	Processor translating ROI to pulse signal. LiCvpr currently only returns mean green value - choices: {LiCvpr,Pos,Chrom}
--winsize	32	Window sized used in some processors (in frames)
--bandpass	0.5,2	bandpass frequencies for processor output
--blobsize	150	quadratic blob size of DNN Face Detector
--draw-facemark	False	draw landmarks when using facemesh detector
--blur	-1	pixelation size of detected ROI
--video	0	video input device number or filename
--limitfps	None	force a delay when reading frames (specified in milliseconds)
--savepath	''	store generated signals as data frame to disk

Camera setup

For rPPG to work as intended it might be required to tweak the camera settings. For example auto white-balancing and auto-exposure could be disabled, to avoid small adjustments in RGB values. One possibility to change the camera settings is via the ffmpeg commandline tool. See the discussion here: https://superuser.com/questions/1287366/open-webcam-settings-dialog-in-windows/1511657

---

---

1. W Verkruysse, L O Svaasand and J S Nelson. Remote plethysmographic imaging using ambient light. Optics Express. 2008;16(26):21434–21445. doi:`10.1364/oe.16.021434 <https://doi.org/10.1364/oe.16.021434>`_↩

2. Li, X., Chen, J., Zhao, G., &#38; Pietikainen, M. (2014). Remote Heart Rate Measurement From Face Videos Under Realistic Situations. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 4264-4271.↩

yarppg's Issues

Live digital filters as a separate package / library

Hello,

I read with interest your blog post https://www.samproell.io/posts/yarppg/yarppg-live-digital-filter/

I wonder if a separate package just for implementing live digital filters in Python won't make sense.

It could be based on list as you did in

yarppg/yarppg/rppg/filters.py

Line 15 in 613155c

self._xs = [0]*len(b)

or on deque as mentioned in your blog post.

A last idea could be to rely on circular buffer / ring buffer on top of Numpy such as https://github.com/eric-wieser/numpy_ringbuffer or https://github.com/Dennis-van-Gils/python-dvg-ringbuffer

Any opinion?

Video Input

Input video as a recorded video does not work. The error is "No frames received"