Human Detection by Boosting-Based Co-occurrence of HOG and Color Feature

Abstract

In this paper, we propose a method for human detection using co-occurrence of Histograms of Oriented Gradients (HOG) features and color features. This method expresses the co-occurrence between HOG and color features by Adaboost and generates the combination of the features effective for the identification automatically. Color features were calculated by making histograms that quantized hue and saturation in local areas. We show the effectiveness of the proposed method by identification experiments for human and non-human images.