|
Our team got a fairly accurate cascade classifier for tracking boulders as well. It also uses depth as well for Boulder tracking as I'm not certain on what strategy I plan to employ. I trained on 900 manually classified images; a competition ready program probably needs even more than that for precision. I can provide the parameters I used to create the datasets later if that will help. There are a few peculiarities one needs to account for, such as the training program's requirement of more images beyond what one specifies with -numPos. I have a few programs that I used to create the sample as well.
|