Traincascade is stuck over 3 weeks

edit

It can't find negative samples. That means that you should either choose a smaller number of negatives (-numNeg), or provide a richer set of negative images.

Also the number of positive samples is probably too high. I would suggest you to start from let's say 2000 for both positives and negatives and then try to increase both of them. But IMHO 5000 positives per stage is too many, you need to have some special reason to set such large number. Otherwise it may result in overfitting...

You don't have to wait for the training to finish. It might already be a good enough classificator. If it doesn't deliver good results, you should start a new training with a different positive/negative ratio, as Kirill suggested.