That seems to work a little better for me in terms of accuracy. This approach is different in that this is using a fixed eye position and overlay measure on the camera image and then working out the arctangent etc. Clever.