Design of Algorithms for People Segmentation Using in Scenes for Non-Invasive Monitoring of Vital Signs Through Video

EasyChair Preprint 8044

Abstract

This project is the result of multiple investigations related to technologies that allow manipulation of the elements present on stage. These investigations provided a broader overview regarding the possible use of technologies and allowed
selecting the tools that provide greater benefit and adaptability to future changes in the project. This implied that a new system was developed, which allows detecting the faces present in a video from a file or a video in real time and later segmenting and saving each face detected in an individual video, for this two algorithms are used, one of them implementing an artificial intelligence approach and the other a traditional image processing approach in order to compare them at the end of development and thus determine which of the two is more optimal in terms of use
of computational resources for the execution of the system in question. Each one is analyzed taking into account the least amount of resources used, to later extract the vital signs per person using Eulerian Magnification. With this project, vital signs monitoring systems will benefit, since large and complex instruments that must be connected to the body of patients that may bother them or limit their care will no longer be required, since now the entire monitoring process of the vital signs will be performed only by analyzing the images of a video using the Euler Magnification algorithm, in this way a non-invasive monitoring is guaranteed that avoids having to connect devices
to the patient’s body.

Keyphrases: Artificial Intelligence, Eulerian motion magnification., Traditional image processing, vital signs

@booklet{EasyChair:8044,
  author    = {Gabriel Brenes-Vega and Oscar González-Alfaro and Luis Chavarría-Zamora},
  title     = {Design of Algorithms for People Segmentation Using in Scenes for Non-Invasive Monitoring of Vital Signs Through Video},
  howpublished = {EasyChair Preprint 8044},
  year      = {EasyChair, 2022}}