基于密度分类及组合特征的人数估计算法*

为了克服不同人群密度及所采用特征对人数估计的影响，提出了一种基于人群密度分类及组合特征的人数统计算法。该算法包括离线特征组合选取和在线实时估计两个阶段。在离线阶段，选取密度阈值将图像样本分为高、低密度两类，然后通过实验方法选取最优的特征组合。在线估计阶段首先通过分类器将样本分为高、低密度两类，然后利用离线阶段选取的特征组合训练得到高斯模型，并分别对两类样本进行人数估计。实验结果表明，与不分高低密度相比，平均估计误差由10.6%降至8.1%；与目前主流的人数估计算法相比，本文算法的平均估计误差也更小。

Crowd counting based on density classification and combination features

The article proposes a novel crowd counting algorithm based on density classification and combination features which can overcome the influence of crowd density and features utilized in crowd counting applications. This algorithm includes two phases: offline combination features selection and on-line real-time estimation. In the offline phase, we use a density threshold to classify the image samples into two categories. Then, the experiments determine the optimal combination features. In the on-line phase, firstly, we classify all images into high and low density categories using a classifier. Then, we use the Gaussian model to train the selected features in the offline phase, and estimate the two groups of images, respectively. Comparing with the method without considering the density difference, the experimental results show that our method can decrease the mean relative error from 10.6% to 8.1% and the proposed algorithm outperforms the state-of-the-art crowd counting methods in average estimation errors.