Limelight doesn’t do super well with color recognition, would anyone have ideas for other ways to tell the difference between gamepieces?
The main way seems to be training a neural network using Google Coral.
can also recomend using a pixy cam for color based tracking. not nearly as advanced as a limelight but it does the job very well
I have not played around with limelight admittedly, but a neutral network is a hilarious case of using a sledge hammer to drive a tack. Yellow and purple are pretty distinct, basic HSV thresholding should do the trick for a ton of applications.
The limelight has a pretrained model that already works well out of the box and so is faster to setup than thresholding is.
The main advantage is being able to detect the orientation of the cube, and that allows for automated intaking if your robot requires a certain orientation of cone.