Usage of HoG (histograms of oriented gradients) features for victim detection at disaster areas

Employing robot teams at disaster areas requires usage of autonomous navigation methods. Moreover, autonomous navigation requires autonomous victim detection. Human skin color based victim detection methods may not be robust due to the variations in lightening conditions at disaster areas. Histograms of Oriented Gradients (HoG) were presented as an alternative way of human detection. In literature, HoG based methods proved their efficiency on the datasets including upright humans. But, the victims have very large variation of poses at a disaster area. In this work, the efficiency of HoG based methods was investigated on a dataset including very different poses and lightening conditions. We have reached 95% success on automatic victim detection problem in real time simulation environment.