|
Target Localization Module?
Anonymous |
11 year
|
So I was browsing the FRC tutorials and was reading Vision Targeting #2 which talked about a target localization module. However when I opened up roborealm, I could not find it. How exactly is this done?
http://www.roborealm.com/FRC2013/slide050.php
|
|
|
|
Steven Gentner from United States [1446 posts] |
11 year
|
Its not actually in RR just yet. We have the module ready but have delayed its release for a couple reasons.
1. Its a very sophisticated module that requires the camera be calibrated and the target to be very well illuminated (i.e. Kinect or really bright LED light). If you are using LED's the 3 ring solution is needed (i.e. 3 rings inside each other).
2. It requires a decent amount of computing (you're not going to be able to run it on the Classmate). Quad core would be ideal.
3. We're not sure if teams are really interested or ready for such a module. It provides X and Y coordinates of where you are in the field (in inches) assuming the camera is looking some part of the target. For constant localization the camera would need to pan around when the robot turns (unless you are using mecanum wheels) which is probably more than what most teams have time for. Naturally, the further you are from the target the more noisy the position is ...
The advantages are that it is a much better detector than just looking for a square (you know where each target it) even if a lot of stuff is in front of the target (i.e. the pyramid). In vision, the more context you take into account the better identification you can perform, that's what this module is designed to do. Its an ideal solution for the autonomous period if you really want to move your robot around the field before getting manual control.
So we're look for a couple teams that may be interested in trying out this technology (its a brand new module) that may want to try out a much more sophisticated vision detection.
Is that something your team would be interested in?
STeven.
|
|
|
|
Anonymous |
11 year
|
The reason why I was interested was because the blob filter module just doesn't seem to be able to filter out the top goal. The only thing that seems to work but not always is aspect ratio. Using the top filter works only if the robot is in front of the goal, but it doesn't work from the sides. Is there any other way to filter out all the targets but the top more effectively?
By using this target localization, I was hoping to produce better results since the robot would know exactly where it is. Also, like you mentioned, it would be very helpful in autonomous.
If the Target Localization isn't that difficult to set up, I wouldn't mind using it at all. The only problem is that our camera only has two LED rings and our driver station laptop might be under powered.
|
|
|
|
Steven Gentner from United States [1446 posts] |
11 year
|
Can you post a couple (at least 3) images of where the top filter fails along with your current robofile (saved RR config)? We will see what we can do to fix that issue based on what you have. We can also run it through the target localization module and see if the light is going to be bright enough. We can work it from there ...
If the TL module works, we can step you through the process.
STeven.
|
|
|
|
Anonymous |
11 year
|
the top filter fails because the other goal is larger and closer to the camera. For example, this image (I don't have access to the robot currently): http://www.roborealm.com/FRC2013/target_4.jpg
I've attacked the .robo file that we are currently running which does not have the top filter and uses the aspect filter instead.
I probably won't be able to get those images for a few days because of midterms :(
 program.robo
|
|
|
|
David from United States [3 posts] |
11 year
|
|
I am very interested in this module. I've been checking the past couple days to ascertain its status. We are running an i7, and have just ordered brighter LED rings which are so bright that it has to seriously adjust the exposure when seeing the target at 10 feet or less. So, in essence, we fit the bill of who this module would pertain to and would be very willing to try it.
|
|
|
|
Steven Gentner from United States [1446 posts] |
11 year
|
David, which LEDs did you order? Seems like you'll have a good setup for the TL module.
To all those interested we've updated that tutorial to include a requirements section. Please take a look over that section as it outlines possible configurations and also describes building the target which will be needed for testing.
We will get the module available in the next couple of days (lots more documentation to do) so if you don't already have a target you can use (as in a REAL target ... not just one square like we see all over YouTube) start building one ... you'll need it! Like David, you may need to order LEDs or more retro tape (see link in tutorial for that too).
http://www.roborealm.com/FRC2013/slide050.php
More to come ...
STeven.
|
|
|
|
David from United States [3 posts] |
11 year
|
|
|
|
|
Steven Gentner from United States [1446 posts] |
11 year
|
David,
The Target Localization module has launched. At the bottom of
http://www.roborealm.com/FRC2013/slide050.php
we've include a zip file that has all the modules in it to start testing (assuming Axis camera).
You'll need to download RR again using your original link to get this new module.
STeven.
|
|
|
|
David from United States [3 posts] |
11 year
|
When testing, the network tables seem to slow down fps the most. We can have 30 fps but when network tables are enabled it drops down by at least 20 fps. It's not even so much the programs that slow it down. I don't see the actual output values consuming so much bandwidth, so is this to the fault of the module? Of all things, it doesn't seem like the network tables should slow us down the most.
Thanks for your help.
|
|
|
|
Steven Gentner from United States [1446 posts] |
11 year
|
David,
Agreed, in theory the NT module should not reduce the fps at all. We did notice a couple places where the overlap may have caused this and updated the module to reduce this dependency.
Please download 2.49.2 for this update.
We didn't notice any reduction of fps when directly connected via ethernet so the issue may still be present.
One thing to check is if the connection is being terminated and restarted many times ... we have a current report of this happening in the C++ version of the Network Tables on the CRio. Perhaps you are using the same platform?
STeven.
|
|
This Thread