An implementation of the paper published by Piotr Dollár, et al: Integral Channel Features
This implementation was tested on the INRIA data set with not very accurate results on the testing data, there are pending improvements for speed and training for this implementation.
My findings and implementation details are described here: Spanish version