Zenithal people counting using histogram of oriented gradients

There are extensive literatures on pedestrian detection, mostly supposed that visible humans are observed in flat view. Sometimes pedestrian detection from another perspective should be considered. In this paper, the pedestrian detection with a vertical view camera is taken into account, which is usually used to count the pedestrian number. The target characteristic is quite different from traditional one though their targets seem the same. The pedestrian head in vertical view is remarkable and a histogram of oriented gradients (HOG) descriptor is introduced for the head detection. The proposed descriptor originates from traditional pedestrian detection and it is reformed and adaptive in our task. To our knowledge, this descriptor is firstly applied to detect pedestrian heads in vertical view. Experiments show that it is effective, and can be applied to count pedestrians in vertical view with a faster processing speed.