There is plenty of technique in 3D reconstruction (shape from X, sapce caving, voxel coloring, structured light, etc.). The "simplest" way to do it from your matched data and your calibrated cameras is to triangulate i.e. compute the intersection of rays emit from 3D points. But this way to do is sparse and usually noisy. Have a taste with voxel coloring, it is fun but not necessarily the most accurate or the fatest technique.

Once you have a P = [K [R|t]] for each camera for each correspondence you have to triangulate in order to have a 3D point. => i.e attachment

You will find code and examples here: http://imagine.enpc.fr/~moulonp/openMVG/

I think to this example (Essential Matrix + triangulation)

https://github.com/openMVG/openMVG/tree/master/src/openMVG_Samples/robust_essential

In order to have good quality 3D points your camera must have a not null baseline.

http://imagine.enpc.fr/~moulonp/openMVG/Triangulation_n_geometry.png

You can refer this book:

http://www.robots.ox.ac.uk/~vgg/hzbook/

Also refer work of Andrew Davison.

please see if the www.openptv.net helps. disclaimer: I'm one of the developers.

I think that the best method to triangulate the image coordinates to 3D Scene is to utilize the toolbox of vision machine in Matlab2019b,

best regards.