09-04-2015 08:03 AM
It's experience-related question about cameras and image processing.
I have a laboratory setup. Under some simplifications it looks like a red ball rolling on a white table (the real setup is shown here https://www.youtube.com/watch?v=kyvW5sOcZHU). A computer vision system should measure X and Y positions of the ball.
For now I use a color camera (with Bayered output) and I find the ball by applying a simple color filter (InRange for R, G, and B components) to the frames. It works, but I would like to increase precision of the measurements.
Question:
Would accuracy of measurements be better in a case I used a mono camera with the same parameters as bayer-camera?
I guess it should be so in a perfect case, i.e. when the ball is definitely highlighted on the frames. But in reality brightness filter usually works worse than color filter. On the other side I can change the shooting conditions, because this is laboratory setup. I can change the color of the objects, I can change lighting.
09-04-2015 09:10 AM
Theoretically i think it is more precise, due to the absence of the debayering filter. I think it's better to choose complementary colors for the ball and the background (black on white?).
09-04-2015 09:34 AM
Edges are a lot fuzzier with color cameras due to bayer filtering, but I suspect finding a large circular object works pretty well due to the large number of edge pixels. Theorectically, a grayscale image would work better because the edges are more clearly defined. When using color, I would probably use the hue and saturation images with any color other than red. Red's hue is zero, which is the same as grayscale.
Bruce
09-04-2015 09:58 AM
Thank you guys for the answers. According to your comments it's better to use grayscaled camera, and shooting conditions should be as good as possible.
I thought about drawing the ball by fluorescence paint. I can install ultraviolet lighting so that the only ball re-emits this light. So, the camera can see the only re-emitted light from the ball and reflected from other objects this re-emitted light. The ball should be much brighter than other objects. What do you think about this?
09-09-2015 09:18 AM
Probably overly complex. I would stick with one of two approaches. Black/white contrast between ball and background, or paint the ball with a color with a significantly different wavelength than the background, and use a color filter on the lens so you only see the ball. Don't use a white background in this case, because part of the white wavelength matches the ball color.
Bruce