Is there a rough rule of thumb we should be following in terms of achieving the best positional accuracy? I'm guessing that you get the best result when the project and camera focus are at their sharpest, and that if scanning a significantly large object, that the loss in angular resolution is made up by better focus across the imaging area by having the scanned object moved further away?
Hi Joseph,
Great question.
If you want a rule of thumb for accuracy, then it should be the closer the better.
But I will caveat this with the fact that most scans do not need the best accuracy, or the highest resolution. Side note: If anyone reading this is scanning on high, you should probably stop, it just adds processing time for very little benefit.
Anyway, for those who are interested in a bit of a deeper dive, read on:
As you pointed out, there is an interplay between image sharpness, projector sharpness, stereo camera geometry and scanning distance with respect to resolution and accuracy. I'll try to descibe the affects of each one independently. Let's start with the easiest one...
Scanning Distance:
Resolution - Due to perspective, the closer the object is to the camera the more space it takes up in the image, and farther away it is the less of the image space it takes up. Since scan resolution is basically (for the sake of this conversation) mapped directly to pixels, the closer your object is, the more scanning resolution it will have.
Accuracy - The same is true for accuracy. But this is mostly tied to the geometry of the two cameras. It is safe to say however, that objects that are farther away are more difficult to measure accurately.
Camera Geometry:
Resolution - The geometry has no affect on resolution, that's strictly distance. However, you can't scan what you can't see. The scanner requires both cameras to see the same spot to capture a point, so a scanner with a cameras that have a large angle between them will produce less points because they can't see as much overlap. For example, imaging two cameras, that are directly beside each other; Both cameras will see almost the identical image. Now imagine a pair of cameras that are 90 degrees from each other; There is very little overlap in the objects that they can see. Now imagine the worst case, where a pair of cameras are 180 degrees from each other (facing each other); There will be no overlap on the object as one camera sees the front of the object, and the other camera sees the back.
Accuracy - This is the opposite of Resolution, where geometry has a very strong affect on accuracy. Imagine the same example as above, two cameras at 90 degrees from one another. Now imagine a ray from each camera, crossing over the same area, meeting at the exact point on the object you're scanning. However, it's best to imagine this intersection point as a fuzzy ball, like an electron cloud, where the surface of the object your scanning is somewhere in that fuzzy ball. In the 90 degree scenario, the fuzzy ball will be the smallest, and so the accuracy is at it's best. This is because one camera is measuring along one plane (X,Y), and the other is measuring along a perpindicular plane (Y,Z). The intersections between those fuzzy areas is small.
Now imagine the cameras are very close together. The fuzzy intersection point would be stretched along the rays of the camera until they stop intersecting. That makes the fuzzy area long and narrow. This would result in the accuracy in the depth of that point, to be the worst case scenario. And if the cameras occupied the exact same space (physically impossible), there would be no solution at all.
Projector Sharpness:
The projector acts like a texture painter. It coats the surface of the object your scanning in a design that can be read by the cameras. However the algorithm we designed is mostly agnostic to the focus of the projector. If you look close at the patterns that are projected, you'll notice the lines aren't sharp. That's intentional, and it means that a slightly out of focus projector is actually beneficial, as perfectly sharp projectors will show individual pixels. In this way, we are completely untethered to the resolution of the projector.
There is one limit of course, and thats if the projector is very out of focus. In this scenario, if the scanner projects a high frequency pattern, it could be completly blurred which would cause missing areas of the scan.
Image Sharpness:
Resolution - If the image is blurry, you could loose points (as exampled above in Projector Sharpness), but otherwise does not affect resolution. A blurry image still contains the same number of pixels as a sharp image.
Accuracy - On the other hand, accuracy is strictly connected to image sharpness. A simple example is to imagine the fine details on an object. A blurry image would "smooth" out the details, making the accuracy bad.
Hopefully this gives you some insight when making scanning decisions!
