Pixy - Axis Camera Alternative?

[Iaquinto.Joe]

http://www.kickstarter.com/projects/...-vision-sensor

Pixy is a camera that connect to the cRio through a master-slave configuration with an arduino. It can also use many other signals like i2C and even analog and digital.

Is this camera FRC legal if bought off the kickstarter?

Would this camera make a superior replacement for the Axis Cameras? (I think it would )

[techhelpbb]

Another question is whether this will work well with retroreflective tape or only bright colors.

[Iaquinto.Joe]

Originally Posted by techhelpbb

Another question is whether this will work well with retroreflective tape or only bright colors.

If you arrange your LEDs as the left half one colour, and the right half a different colour, it could show up as a colour tag. I feel like the camera would pick it up well, considering the Axis make the colour pretty visible on the direct feed.

[Joe Ross]

Originally Posted by Iaquinto.Joe

Is this camera FRC legal if bought off the kickstarter?

I don't believe that the limited availability through kickstarter meets FIRST's definition of a VENDOR, and would not be legal. However, if you purchased it directly when it is generally available, that would be legal.

Originally Posted by Iaquinto.Joe

If you arrange your LEDs as the left half one colour, and the right half a different colour, it could show up as a colour tag. I feel like the camera would pick it up well, considering the Axis make the colour pretty visible on the direct feed.

You would need some type of blinders for this to have a chance of working.

[Foster]

I ordered one with the pan/tilt option. So in a few weeks I'll be able to report back on how it does on distance and refecting tape.

Based on how they are doing this, it should be a COTS product by January. Their ship date is November, so once they ship the present order for 2,500 they should be good to go.

[Gdeaver]

This is an evolution of the Cmucam. It's blob detection. Using Opencv and NI vision can give true object detection and more. Trade off of cost and processing power. This could have some nice application on robots.

[Foster]

Yes, it's blobs, but one robots blob is another robots object.

I'm looking at the off board processing. They are doing most of the heavy lifting on the camera processor and sending it down to the host for further processing. So the overall load is less, which frees up CPU cycles for other processing. Does that open up a chance for two cameras to get more exact range information?

It's also not as demanding to know OpenCV, so teams with less hard core programming resources may be able to use this easier.

Anyway, it's a while before I get one, but I'm excited about the possibilities!

[techhelpbb]

Originally Posted by Foster

It's also not as demanding to know OpenCV, so teams with less hard core programming resources may be able to use this easier.

I agree this could be a really cool choice if it works out.

However OpenCV in Python on a laptop on the robot is really not hard to use.
Many working examples exist.

[theprgramerdude]

The LPC4300 used on the new CMUcam is more or less on par with the power in the Crio, so it would be viable to push all processing onto the camera board. The only limitation is the low internal memory which can't buffer a single frame at full resolution for the price they probably could have thrown an SDRAM chip or PSRAM on their for a little extra.

[Gdeaver]

Hummm. I wonder what you would get if you pointed that straight up at the ceiling. Maybe a little adjustment to the camera settings, a little code tweak. It maybe almost there. I'm talking about the way our ancestors navigated.

[Foster]

Originally Posted by Gdeaver

It maybe almost there. I'm talking about the way our ancestors navigated.

Esso roadmaps?

[gpetilli]

Our team just ordered a couple of these cameras for next season. Take a look !!!

http://www.kickstarter.com/projects/...-vision-sensor

50 frames per sec of processed target XY coordinates - no need to transmit the image data over saturated WiFi links!!! Simple I2C interface.

[faust1706]

Originally Posted by gpetilli

50 frames per sec of processed target XY coordinates - no need to transmit the image data over saturated WiFi links!!! Simple I2C interface.

just wondering about this camera, err....camera and software I guess. It seems all fine, but the question still remains, how does it act in different environments? The problem with using colours is that the difference between different rooms in the same building, or even different parts of a room, dramatically change the value of the object.

[gpetilli]

The hue algorithm they brag about is intended to remove the intensity variations. My expectation is that if you train it with an LED ring light, it will have a very strong affinity to your LED color. This should play well with the retro-reflective tape FRC uses and should eliminate effects of ambient lights. You may want to pick a unique LED color for your ring light so you don't lock onto another robot thinking it is a target - unless that is what you want?
Obviously no one has tried this on an FRC field yet. Given it is available to everyone, I cant see any reason it would not be legal. If they run out (which they have not yet), it is no different than the year that FRC consumed the world supply of fisher price motors and many teams were left wanting. We ordered two for $115 and will post how they work out when we get them. Not much of a risk and potentially a big advancement if they live up to expectations.

[faust1706]

Originally Posted by gpetilli

The hue algorithm they brag about is intended to remove the intensity variations. My expectation is that if you train it with an LED ring light, it will have a very strong affinity to your LED color. This should play well with the retro-reflective tape FRC uses and should eliminate effects of ambient lights. You may want to pick a unique LED color for your ring light so you don't lock onto another robot thinking it is a target - unless that is what you want?
Obviously no one has tried this on an FRC field yet. Given it is available to everyone, I cant see any reason it would not be legal. If they run out (which they have not yet), it is no different than the year that FRC consumed the world supply of fisher price motors and many teams were left wanting. We ordered two for $115 and will post how they work out when we get them. Not much of a risk and potentially a big advancement if they live up to expectations.

we use ir light because a) the kinect has an ir camera (and the asus xtion that we will be using from now on) and b) it doesnt cause people to look away if the robot is facing towards you. moving on.

Given how it apparently can track anything, I assume it preforms a blur (perhaps gaussian) then thresholds. This could work in multiple enviroments, but could you teach it to track a white square with whatever is inside it(chains, at the terra heute regional, sunlight went through windows, through the 3pt goal and nearly blinded us, but due to a failsafe check, it didnt.) The issue is that the target is not static, it change.

I can see a lot of teams using this for piece pick up (autonomously?). We had both sides of this working for frisbees, crio and vision, just didnt have a method of picking them up XD. I still think teams will stay with the axis camera due to familiarity, and other, more courageous teams, will attempt to use pixy.