This is my very first project in the field of computer vision. Basically, I tried to recreate the OpenCV PeopleDetector, by training the SVM manually instead of using the pre trained one of OpenCV.
The training dataset in the main.py file is composed by the INRIA and VIPER datasets, joined together. For what concerns the negative examples, you can find the procedure in the utils.py code. The test was performed on the CROWD_PETS_2009 dataset, but you are free to use your own.