as far as I know they are no tools that computes ROC curve in OTB, but the training applcations (
TrainVectorClassifer) output confusion matrices computed using the validation sample (or the training sample if there is no validation sample available). It is also possible to compute a confusion matrix from raster classification and ground truth with the
You can then deduce the ROC curve at different thresholds. But which threshold do you want to use with the random forest classifier ?