Pruning the density estimator

1000 bins         or less!

Reduce the number of bins through clustering

Result: Detection/Classification is faster than template correlation