People Detection in Complex Scenes By Using An Improved and Robust HOG Descriptor

Abstract

The detection of moving people in a complex scene filmed with a single camera is among the most difficult fields of research in vision by computer. In this work, we suggest improving the quality of detection methods based on the histogram of oriented gradients (HOG) descriptor. For that, we purpose to use a combination of type detector/descriptor to minimize the rate of the false detection produced by the descriptor HOG. The implementation of this combination as well as its evaluation on public bases show clearly that the technique which we propose produces many good results at the level of the detection of the people in movement compared with the descriptor HOG