Which color are you tracking with your Camera?

Hi guys,

I’m doing this poll in order to share ideas and obtained results about the best color of the target for the tracking system.

How we know, the backboards has a retro reflective tape and we need to point a light in them in order to distinguish better what is the target.

Up to this moment, we just tried to use GREEN lights around the camera’s lens and we’ve obtained a very good calibration enviroment.

Today, we’ll try another colors (Red and Blue), but we’re scared about luminous interference from the field.

What you guys have discussed about this?

We’re planning on testing a few different options, but leaning towards solid white (with a grayscale image, extract luminance plane, threshold) or IR.

Make sure you check out this article before you decide on white… it seems to make a good argument against it! http://www.chiefdelphi.com/media/papers/2620

Could you please quote the relevant text? I’ve read that whitepaper and skimmed it a few times over, don’t see anything on the disadvantages of using white LEDs.

The beauty of retro-reflective tape is, what ever color you choose to use, you see that color. Now, there is a mild chance that another team may be directly in line with you, or near enough to you, and their light may pollute your color a bit. So, just make sure that whatever color you choose, make sure your calibration allows for slight variations to “your color”.
That being said, has anyone measured the “usable angle” that the light is reflected back with? In other words, how far apart can the light be from the lens and still have a reliable color return?

I think a good amount of teams will be using green because the light ring on FIRST Choice is green.

That is what we will be using.

Look at the pictures on page 6.

With white, you can perform the initial filter/mask, then your convex hull operation. While this identifies the targets, it also identifies a bunch of other “stuff” - lights, reflections from the plexiglass, even the basketball nets. That makes this operation difficult to use, since you then have to go on to filter our the blobs you don’t want. There are other ways to identify what you want (like the edge detection mentioned later), but in general they’re going to take more processing power to work, from my experience.

With another color, you don’t have that problem. You can perform your original filter/mask, then a convex hull operation. Now you’re left with only your targets showing up, and no extraneous “stuff” you have to filter out - take the center of each blob and your done!

You just need to watch out for that pesky team at your regional wearing bright green shirts… you don’t want to detect them!

I guess that the only problem that you could have using white leds is about the other lights into the field.

The lights above the field and the reflex provided by the alliance’s wall could return erroneous measurements when you’ll calculate the size of the target.

Remembering that, the enviroment that we play the match is very different of the enviroment that we build the program.

Remember that the field has red, blue, and amber LEDs at each end if the field where the alliance stations are. There will be less “noise” using the green LEDs so that makes them the mos obvious choice.

True, we’re waiting to see how the imaqDetectRectangles call performs on the cRio…just gotta get our backboard and ring light manufactured.

We are thinking of using two color of lights taking two pictures one with each light. Then the tape should be the only thing that changes color between the two frames.The frame rate will be slower but data for figuring out what is the tape should be cleaner.