With two cameras in a calibrated setup this is pretty simple. You’ll just have to shoot rays from the camera positions through the keypoints on the view plane and calculate the line-line-intersections. This returns the closest points on the lines which you can average.
2 Likes
Thats still cool 
nicely done.
However what software do you use?for the basic photo pose detection? I tried installing the packages on the PC, but I can’t seem to manage it. Is there a go to way? What additional package would I need to install? Because as i tried it kept asking for packages. The again maybe visualStudio might not be the best platform for it . Might sound silly, but I’m at a learning phase when it comes to scripting.
P.S. found this one. might be interesting as well http://gvv.mpi-inf.mpg.de/projects/VNect/
1 Like
Yeah I think this monocular detection is also based on PoseNet. Of course, if you have a bone model then you can also use a solver which will try to make it fit to these keypoint lines. With well defined limits for the bones this seems to estimate the poses pretty well (and also fast, at least with C++). But thanks for the hint, I’ll watch out if there is something available for javascript…
I had a closer look at posenet because I was already using another tensorflow library. But you’ll have to find your own way using it with your preferred development environment. There are good tutorials online. I’d start here: https://www.tensorflow.org/install
Edit: But probably you do not even need that. Maybe npmjs then node.js and posenet will already work…