Aerial Camera for FIRST matches

[faust1706]

If FIRST would add an aerial camera and give the feed to each team playing, I believe it would greatly increase the level of competition, given that teams actually use it. My team has been working on the software for an autonomous robot for 2 years now, and this problem could be solved in A WEEK if FIRST would simply add an aerial camera. I hope one day this will be a reality.

[MrTechCenter]

Originally Posted by faust1706

If FIRST would add an aerial camera and give the feed to each team playing, I believe it would greatly increase the level of competition, given that teams actually use it. My team has been working on the software for an autonomous robot for 2 years now, and this problem could be solved in A WEEK if FIRST would simply add an aerial camera. I hope one day this will be a reality.

Way too soon for this to happen. While it would certainly be helpful, it's just not going to happen anytime soon. Although, at the MadTown Throwdown last year, Team 1671 the Bird Brains developed a web app that allowed coaches to have a tablet and see the live score as well as the live stream.

[pwnageNick]

I know for the robocup competition, teams are required to have certain labels and symbols on the top of their robot down to a spec, and they have a top view camera that teams can access during the match to track where players are and do autonomous programming off of that. This would be a very cool thing to have in FRC. While it would take some time (years) for teams to fully utilize it, it could make autonomous mode a lot more interesting and have more action between alliances.

Think about the defense we saw on Einstein and at IRI with the goalie bots such as 1114 against/with 254. Now think about the crazier chess matches that would ensue if both robots knew where every robot was on the field and tried to outmaneuver the other. And that would be without a human driving it using a kinect, which I know some people are against being legal.

I don't see this happening for this coming season, or even the one after, but it is a very intriguing thought for the future.

-Nick

[Greg McKaskle]

The biggest complication with this is that teams would probably not have access to this until an event. It would probably be pretty hard to replicate it in a build space, and if you did, there would probably be a number of significant difference.

We've done this a number of times with NXT robots or smaller and slower robots playing soccer or other games. It is still a pretty difficult challenge, and you really want robots wear consistent markers that the camera can view -- robots with hats.

If you believe that one mounted camera and one week is all you need, you may want to investigate where your team can mount cameras. What did the rules say last year?

Greg McKaskle

[faust1706]

Originally Posted by Greg McKaskle

If you believe that one mounted camera and one week is all you need, you may want to investigate where your team can mount cameras. What did the rules say last year?

Greg McKaskle

I was thinking that collecting data at the first regional we go to, then play around with it. I wonder if we could get a quadcopter to give us a camera feed like this team did just to simply video tape the match...

https://www.youtube.com/watch?v=wCDUrJ4M6pk

[Andrew Schreiber]

Originally Posted by faust1706

I was thinking that collecting data at the first regional we go to, then play around with it. I wonder if we could get a quadcopter to give us a camera feed like this team did just to simply video tape the match...

https://www.youtube.com/watch?v=wCDUrJ4M6pk

The less things flying over my head at events the better. The last three years have had me watching for things flying at my head... I don't want to have to worry about something dropping ON me.

[BBray_T1296]

Quadcopters in crouded rooms/arena floors is really a no-go. I know the stuff is relatively reliable but it makes a lot of people (even myself) worried when we see a chopper with no blade guards hovering over the crowded stands.

^This happened at TRR for a little bit

I am all for the versatility and capability of quadcopters/etc, but it only has to come down once to be a major problem.

Also FTC events have major problems with other unencrypted 2.4GHz stuff is going on in the same roof. Their system does not handle bandwidth pollution hardly at all before faulting.

[yash101]

Using an aerial view, tracking all the field targets would be just as simple as background substraction. However, I have great doubts that FIRST will actually do this. However, there is a way how you can simulate this. Put a camera at the driver station window, as high as possible. Use a suction cup to keep it in place. If done right, the camera should be above most of the robots. Now, you can apply a perspective transform to make the image look like an aerial view. This would look quite close to an aerial view. Of course, you are using a perspective transform and transforming nearly 90 degrees so there will be a heavy loss of resolution, but that shouldn't cause any problems. Just use a 1080p camera. All the processing is quite easy, so it wouldn't hurt to use a high resolution camera. After the perspective transform, you can scale it down for your actual tracking!

Originally Posted by BBray_T1296

Quadcopters in crouded rooms/arena floors is really a no-go. I know the stuff is relatively reliable but it makes a lot of people (even myself) worried when we see a chopper with no blade guards hovering over the crowded stands.

^This happened at TRR for a little bit

I am all for the versatility and capability of quadcopters/etc, but it only has to come down once to be a major problem.

Also FTC events have major problems with other unencrypted 2.4GHz stuff is going on in the same roof. Their system does not handle bandwidth pollution hardly at all before faulting.

I thought of this really cool idea a long time ago that could make quadcopters safe in an application like this -- MAGNETS!!!. A servo could be on a worm-gear drive, lowering and raising a magnet into a separator tube. The magnet is strong enough to lift the entire robot (maybe 2-3KGs). Now, what happens is that during setup, the quadcopter, on which this is mounted will fly up to some steel truss on the ceiling. The servo would be then triggered to move the magnet out, so that the craft can hold onto the truss. The craft then could shut down it's propellers and just idle. For safety measures, an accelerometer could be onboard to alert the craft if it falls, so it can turn on it's propellers and safely land! I am pretty sure that would be extremely safe and a good way to get an aerial view of the field. To make the craft come down, it would first start it's propellers and create enough thrust to keep itself in air. The magnet would then be retracted, causing the craft to detach. The craft can now fly down safely for it to be picked up by field staff and stashed for cleanup

[BBray_T1296]

What if on my robot at the top of the height limit, I place a carpet square that completely hides my robot from said aerial cam. (unfolding to the 20" extension, to mask bumpers too)

Call me the invisible bot.

[MrRoboSteve]

In the "more practical" department, consider this:

1. Put cameras in known locations on field



2. Robots have markers attached.



You get the idea.

3. FMS calculates position estimates for each target and delivers them each cycle to the driver station.

4. In order to make it clear that the FMS position estimates are best efforts, matches are randomly selected to have the position estimates disabled. This forces teams to have a workable strategy in case there are issues with the position estimates.

[Tom Line]

An easy and practical way of doing this at most venues would be to string a high-tension spectra, amsteel, or kevlar line across above the field and put a go pro on a pulley in the middle of it facing downward.

If FIRST would concur to put a 10 foot tall aluminum pole on top of the driverstation at each end of the field to string the line between, you could easily do this at ANY venue.

[EricDrost]

Curie 2014 Match 135 from catwalk:
http://youtu.be/hqoQ5pmK2jI?t=60s

[faust1706]

Originally Posted by MrRoboSteve

In the "more practical" department, consider this:
3. FMS calculates position estimates for each target and delivers them each cycle to the driver station.

4. In order to make it clear that the FMS position estimates are best efforts, matches are randomly selected to have the position estimates disabled. This forces teams to have a workable strategy in case there are issues with the position estimates.

I don't necessarily agree with number 3. They should simply feed the image to the driver station and let them do the heavy lifting of finding the position of the robots. I know the arguments against this ("teams wouldn't have data to play around with before competition"), but all you'd need is an image from all 4 of those cameras with no robots or game pieces in it, then implement a transformation on the image like Yash mentioned to change the perspective. You could even do sterio vision (with homograph (yash)), to calculate distance and use simple trig to calculate distance.

As for number 4, if the fms did number 3, and it sometimes didn't work, it should be counted as a field fault, not an "oh well" situation.

Originally Posted by EricDrost

Curie 2014 Match 135 from catwalk:
http://youtu.be/hqoQ5pmK2jI?t=60s

Hm....that's exactly what I'm talking about.

[Greg McKaskle]

So. You have overhead footage of a match. Plenty of high res pixels to analyze.

The remaining steps are to isolate and track robots. Estimate their heading and velocity. Do the same for game objects. Perhaps superimpose your own graphic robot on top of the image showing how your AI would move your robot.

Sounds like a good project. And as I said in the earlier post, we've done it a few times for demos and it is still a bunch of work and doesn't necessarily work that well.

Greg McKaskle

[faust1706]

Originally Posted by Greg McKaskle

The remaining steps are to isolate and track robots. Estimate their heading and velocity. Do the same for game objects. Perhaps superimpose your own graphic robot on top of the image showing how your AI would move your robot.

All it'd take is a simple background subtraction. Take a calibration image with nothing on the field. Then you grab an image during a match, and delete your calibration image from the image you just grabbed. As for velocity, you simply record the position of your object(s) , find the distance it traveled between frames and divide by frame time.

We do a background subtraction for depth tracking with the kinect (and asus xtion) and it works perfectly.

I don't think knowing the velocity (speed and heading) of another robot would be that useful considering most robots can turn on a dime, and some don't even hsbe to turn to go in a different direction. For game pieces it'd be somewhat useful, but I feel that having a camera above your intake would be more beneficial, but that's just me.