FRC 2020 Vision Tracker

We are wanting to take more experienced suggestions on strategies to solve this years game. This year’s game has 3 sections of retro reflective tape on the bottom half of a hexagon at a bottom height of 8’. We are planning to mount a H7 camera to our turret with a green LED ring light, range finder, and IMU. The most obvious solution is to increase the shutter speed such that the only thing we see is the reflective tape (and maybe ceiling lights) and running that through a hue filter and then find either blobs or rectangles. I see 2 major pitfalls: 1 as our range decreases, at a fixed shutter speed, the tape may over saturate and then the Hue filter will fail. 2 is there an easy way from the camera to project the angle to turn the turret given how the target is moving through the image over time?

Thanks
MechTech FRC 3959



BlueGoal-108in-Center.jpg

Trying Blob with a low exposure and I’m getting the target, the whole image and the lights in the shop. Any idea what I should change?

Hi Jim,

I think you need more resolution than QQVGA. Please use QVGA. Also, while it won’t come out in the next firmware… I will be getting VGA blobs to work (it will be slow, but, it will give you high res). I think your approach on lowering the exposure is correct. That makes a lot of sense to do. With a higher res then you should be able to filter out everything else.

That said… why not IR light? And use the IR less filter lens with very low exposure? This will filter everything out except the tape. IR light from your source hits the tape, bounces back, and since the exposure is really low you only see that tape.

As for your second question… not sure what you mean. Just use the largest blobs centroid?

Thank you so much, you are awesome.

When you say IR less filter lens, you mean a lens that doesn’t have the IR filter on it, correct? Is the filter on the lens shipped with the M7/H7 easily removable? Is strategy to filter out colors and just get IR, or by dropping the exposure, hope that the power of the IR LEDs overcomes other light sources?

Is there a way to do HSL/HSV filtering that I’m missing? I’ve searched the docs with the search tool and just reading through it and couldn’t find anything except LAB, YUV, RGB, and Grayscale.

Is there an easy way (as my experience is non-existent) to use neural Net features or anything for identification of the target and still get a 15-25ms rate?

The rear goal doesn’t have a vision target on it and is 2’ behind the goal with a vision target. So based on range and cross angle, the center of the rear target is low and offset from center.

Again, thank you so much. Last year was the first year we had any real success with vision and that was completely due to you and your products.

Thanks
MechTech

You can remove the IR filter on the lens that comes with the camera or buy our IR lens on the website. Note that without the IR filter working in sunlight will be tough.

The idea is that the IR light when shined in the retro reflector will look extremely bright (all white 255 pixels). Then you just lower the exposure to make sure that only the IR light stands out versus everything else.

Without the IR filter the camera is very sensitive to IR light.

As for HSL filtering… That’s what LAB is for. It’s a color space that has color separated from luminance. YUV/LAB/HSL/HSV are all the same in their use for this case. LAB has the benefit of modeling human perception more however.

Regarding a net, no, unless you know how to train one don’t think about using one.