선택적 블록기반의 HOG-SVM을 이용한 향상된 속도의 보행자 검출

We propose a fast implementation method for the Histogram of Oriented Gradients (HOG) based pedestrian detection. Instead of all the blocks inside the detection window, we extract the HOG features from a few blocks selected by the Adaboost algorithm. Experiment results show that the proposed algorithm is four times faster than the conventional HOG-SVM algorithm.