My team just got a limelight for the first time, and running apriltag detection, we are only getting ~13 fps. To get a reasonable detection distance (10-15 ft) we are using 720p at 1.5 downscaling. 3d doesn’t appear to have a performance impact. I haven’t been able to find any preexisting topics that are up to date on limelight apriltags.
Is this a normal pipeline rate? Any tips on how to make it better?
I think you are way too focused on frame rate. In my experience you want to set resolution as high as possible (I am assuming you have encoders and are doing normal odometry too, if you aren’t you should be). The goal here is less to position constantly entirely off of tags and instead as accurately as possible correct the pose of your odometry. Not saying faster isn’t better but even a few fps is fine.
I can’t remember the frame rate numbers so I will leave that to someone else.
4 Likes
Which Limelight do you have?
3
Based off of playing around with retroreflective pipelines, I assumed I could just pipe the output directly into a control system without factoring in any fancy caching tracking thing. Our team is very bare bones (2 years ago our code was an essentially a straight pipe from controller to motors). I am working on remaking the entire codebase from scratch and attempting to find what is actually going to be important for increasing the productivity of our robot.
We don’t have the technical ability for full autonomous driving like you are suggesting so I guess I’ll crack open VS again and find a way to feed the limelight readings properly into PID controller.
You can absolutely use it like that! I mistakenly assumed you were doing something fancy because of your early and extensive testing.
I would highly suggest borrowing this example code. There is also an example for just aiming if you don’t want it to range.
Regarding frame rate one way you may improve is to carefully analyze where you want to shoot from and cutting off the likely significant area above and/or below using cropping
1 Like