Implementación de un prototipo para la medición del pulso cardíaco mediante procesamiento de vídeo en la tarjeta de desarrollo JETSON TX1

Abstract

Non-invasive techniques for vital signs monitoring have been developing in the last years due to interdisciplinary fields like the computer vision and artificial intelligence. So, in RidgeRun Engineering, there is a need for implementing an image processing system based on the Eulerian video Magnification (EVM) algorithm to detect the heart pulse in the embedded system Jetson TX1, as a way to explore its capabilities. This document describes the design and implementation of this system to detect the heart pulse of a person from video images. The application can be split into two main stages: the implementation of the EVM algorithm in C++ and the post-processing necessary in the video with the magnified signal of interest. In the first stage, the architecture of the algorithm implementation is detailed and for the second stage, the approach to extract the heart pulse is described. A wide variety of tests have been done to verify the implementation, this includes: the calculation of the structural similarity (SSIM), the quantitative comparison of the heart pulse with simultaneous measurements from an oximeter and performance tests. The results demonstrate that the solution, using the GPU, is capable to estimate the heart pulse with an accuracy near the 95 %.