Weighted HOG for Thermal Pedestrian Detection

Pedestrian detection is an important topic in many applications. For the purpose of applications used around the clock, the work for detecting pedestrian based on thermal sensors has attracted significant attentions. Based on the observation that the histograms of four cells in HOG descriptor over the pedestrian contour are significantly different, this study proposes a new feature called weighted HOG (WHOG) that weights the traditional HOG by the difference pattern of the histograms of cells. In the experiment, four videos are used to validat
^e
the proposed WHOG and the results demonstrate that WHOG outperforms HOG in all cases.