Vision Tracking Grey Totes

[btcshields6]

Hey everybody. I posted a few days ago in the general programming section about the 2015 Vision Tracking example. I have yet to get any response, but I have managed to learn a little since then, so I thought I'd sort of restate my question(s).

My goal is to be able to track a grey tote. I'd like to track from the short side (not that that makes a huge difference, as the grey totes don't have tape on the sides). I've looked at the 2015 Vision Example, which includes a VI for shape and color processing. However, I haven't had success with picking up a grey tote.

While I was looking into this problem, I checked out NI Vision Assistant, but I can't seem to find a way to consistently lock onto a tote. I'd like to use the Vision data for centering, so I can't just match the color, which wouldn't consistently hit the same point on a grey tote.

So I suppose I have two questions.

1) How can I get the example to register the grey tote?

and

2) If I can't get the example to run, how can I track a grey tote for centering?

Thanks,

Benjamin Shields

[Greg McKaskle]

I think that you will likely be able to set the S and V values of the target color so that it is pretty good at tracking a gray tote. I'll don't have many images of the gray tote, so I'm not sure what will work for you.

For the images included in the example, Image 18, you can set the Hue to be a wide range, or a bluish range. Set the Saturation to go 0 - 50 and the values to go 0 - 100. That gives a reasonable, but potentially noise mask, but the other metrics tend to clean it up in the images I have.

Feel free to post an image.

Also, while it isn't idea for distant totes, the shape based example intentionally uses high contrast patterns printed on the end of the tote. You can train it for different patterns and adjust the search criteria to get very good tracking once you are somewhat close to the tote.

Greg McKaskle

[btcshields6]

Hmm. Okay. I'll try that tomorrow and report back.

Thank you!

[GeeTwo]

We don't have anything working yet, but our physical pickup is similar to the Greenhorn's Ri3d, but with a "rake" gripper that works like a contour gauge, and a passive "foot". This is optimized for long-side pickup and a bottom-stack, so we're going to try tracking the long side first. We took a few stills with the KOP camera, and using the edge detection algorithms in the latest version of imagemagick to identify those stacking tabs and the corrugations certainly seems promising. We have figured out that we want to get the camera in the 30fps mode, which cuts the screen resolution in half. Even at 640 pixels, though, you can still have 1mm resolution with the whole tote in the field of view.

If we do get this working well enough, we will go for a tote-stack auto routine - drive in rough with the vision targets, line up close with the edge detector, lift, and back up a few feet. Repeat twice, then dash to the auto zone and deliver. We have also purchased some decent range finders (the cheap ultrasonics have a 1" resolution) to get close enough to grab by the lip.

We're also playing with tricks like casting light from different directions, to cast shadows onto the tote.

[TylerStaudigel]

How do you get the default vision codes to work. When we use the default just color code the robot cannot obtain accurate enough values to track the tote, but when we try to use the color and shape vi it can do an excellent job at identifying the tote, but cannot track it relative to center. Finally when we try to use the targets it tracks like ten objects within it's field of view. Anyone got any advice?

[Greg McKaskle]

How do you get the default vision codes to work

Have you followed the tutorial? If you are using your camera with your lighting, it may be necessary to calibrate the color ranges. This is good practice since you will run into this at an event/demo. It is really pretty easy once you understand what H, S, and V mean. Once the mask looks good, the rest is basic trig and math.

Feel free to ask more detailed questions, post images showing the state of failure, and I can give more detailed assistance.

Greg McKaskle

[GeeTwo]

As the totes are grey, it may help to either convert the image to greyscale, so the changes within the tote are relatively exaggerated, OR to calculate color saturation, and know that anything with a lot of saturation is NOT the tote.