by WP2016 » Thu Jun 23, 2016 5:26 pm
This is a great question! I also want to know, especially how the triangulation works in David.
I do not think these regions make any difference in terms of scanning result, the issue should be about the sharpness of the captured projected pattern. This is because the triangulation allows each X1 Y1 of the camera sensor to correlate with X2 Y2 of the projector to calculate X, Y, and Z of the object (the correlation is achieved during calibration).
If my understanding is correct, the issue of accuracy depends on how accurate can the system acquire X1, Y1, and X2, Y2 (sensor recognizing the projector pattern). Once calibrated, the full frame projected area will provide similar, if not same accuracy so far as the object receives a clear structure light pattern and the projected area is within the depth of view of the camera. Thus separating the regions may only be relevant when depth of field is considered.
Based on my understanding (because David's code or algorithm is secret), an optimized scanning for resolution (and accuracy) would require the object filling the view as much as possible (to use as many pixels of the sensor as it can), matching the structure light pattern to the view (to best achieve the pattern resolution, as limited by the projector's own resolution), and setting up a depth of view to enclose the object surface to be scanned (recess surface being the far plane of acceptable circle of confusion, and the nearest surface being the near plane of acceptable COC). But if my presumption was mistaken, such setup would not be necessary.
I am questioning my understanding, because I have seen scanning not tuned to "zoom in" for high resolution and it seems a good result is still produced.
If we can just leave the object to occupy a small portion of the calibration size (say about 20% of width and height of the sensor leaving at the center), then the setup is more versatile and the scan region issue is more relevant. Any comment or clarification is appreciated.
Projector: Acer K132 + 1/2/4/10 X closeup dai. 52 mm.
Camera: Grasshopper 3 GS3-U3-28S4C-C (@ 1600 X 1200, 30 fps); Lens: 8-50 mm F/1.4.
Computer: USB 3.0, CPU: 2.6-3.5 GHz, GPU: 3G NVidia 970M, 32G RAM.