I read the paper, and I am not satisfied that the authors described their methods in sufficient detail for you to use the methods in that particular conference proceeding. It is a conference paper, so is limited in the detail it presents. For example, they do not mention how they arrived at β, the value you want to determine. (I would use a full paper published in a regular issue of a journal, but even that does not mean you will be able to reproduce the authors’ results. I speak from experience, having spent an entire semester not being able to reproduce the results of a particularly promising radial basis function neural network designed to detect and isolate fetal EKG, a couple decades ago, because the authors left out some important details. Neither me nor my professor nor colleagues ever figured out how the authors got the results they published.)
That said, I would use the gradient function to determine the slopes of the QRS complex. Note that there is broadband noise that you may not be able to eliminate by conventional discrete filtering. You may need to use Savitzky-Golay filtering on the original EKG signal to get a smooth-enough signal to analyse using gradient.