As you are not doing stereo, from a single 2D point and camera calibration, you cannot obtain a 3D point (at least no without additional constraints, for example that the point is at the ground level or something like that).
What you can obtain from a single 2D point is a 3D ray (a 3D vector), in which the 3D point lies. Referring to the picture below, from [https://docs.opencv.org/2.4/_images/pinhole_camera_model.png]:

From (X, Y, Z) you can obtain (u, v), but any point in the red line will also project to the same (u, v). This implies that given (u, v), you cannot know which point along this line you are observing.