Ask Your Question

Revision history [back]

Get 3D world coordinates from 2D image coordinates

I have a set of 20 points in a 3D space, and I have a camera as well (along with its intrinsic matrix). I took an image from the camera, and it contained the (x, y) image coordinates of all the 20 points I created a dataset from this consisting of images from various angles/viewpoints, along with the imaged coordinates of all the 20 points(ground-truth). Then I created a neural network that took in an image and predicted the image coordinates of all the 20 points(predicted). Now that I have my predicted coordinates, I wanted to convert them back into 3D coordinates and evaluate my results accordingly.I tried using OpenCV's solvePnP() to get the rotational and translational vectors of my camera but I don't know what to do with this and am not sure whether I'm heading in the right direction?