The simplest answer is: "People can position it within 1 pixel." But that just begs the question: "Well, how much is one pixel?" For the paper I published on my first experiment with the virtual pointer, I had my cameras in the following configuration:
Using the values chosen for the experiment, we find the following:
Pixel Distance Distance % of D for 1 Sep for Pixel for Pixel-1 Pixel Error Sep (D) Sep (D1) (D-D1)/D ---- --------- ----------- ------------- -15 0.821 0.813 0.97% 0 0.954 0.965 1.14% 20 1.259 1.241 1.43% 40 1.810 1.771 2.15% 60 3.212 3.092 3.74% 80 5.237 4.927 5.92%The first column is the separation (measured in pixels) of the left and right images of the virtual pointer as displayed on the monitor.
The second column gives the distances in the remote world (as viewed by the cameras) corresponding to those pixel positions.
The third column represents the distance corresponding to a pixel separation of 1 less than the second column.
The fourth column gives the percentage error of the total distance corresponding to a one pixel underestimation of the position of the object at the camera site.
For the given camera configuration, pixel separations of 80 are getting close to the limits of fusability.
These data can be presented graphically:

where the distance of the object from the cameras is measured in metres, naturally.