关键词: 人数统计, 人群密度估计, 密度分类, 局部特征

Abstract: Since the accuracy of the crowd counting is influenced by crowd density, a novel method for crowd counting is presented. Firstly, in the pre-processing stage, a sub-crowd segmentation method based on sliding window is designed, which improves the efficiency and precision. Secondly, the sub-crowds are divided into high-density and low-density. Then these two sub-crowds separately are trained off-line and choose the best feature combination and the regression model by experimental method. Finally, the selected combination of features and regression model are utilized to predict the number of persons. Compared with the state of the art algorithms, the average estimation error of the proposed algorithm is 18.9 % smaller, which proves the effectiveness of the algorithm.

Key words:  Crowd counting, crowd density estimation, density classification, local features