Abstract: The pedestrian detection mechanism was proposed for solving these defects such as weak expression ability of HOG reduced by automatically ignoring differences in the opposite direction of its gradient direction when taking the current gradient direction histogram as descriptor for applying to pedestrian detection. Basing on analysis of the traditional HOG descriptor, the improved HOG descriptor was designed by connecting histograms; and a new normalization technique was proposed to embed into the improved HOG descriptor for enhancing its expression ability. The experiments results basing on several datasets show that: comparison with traditional HOG feature method, this proposed mechanism had higher accuracy and lower miss rate.

Key words:  , improved HOG feature； pedestrian detection； gradient direction； histogram normalization； lower miss rate