基于尺度空间金字塔网络的人群计数算法

Crowd counting algorithm based on scale space pyramid network

针对人群密度大和分布不均等因素造成计数困难的问题,提出了一种基于尺度空间金字塔网络的人群计数算法。首先通过迁移经典网络VGG16的前10层作为前端模块提取初步特征;然后使用尺度空间金字塔模块提取多尺度人群特征,并在该模块中引入扩张卷积,通过增大感受野进一步优化特征图,以解决人群拥挤和人群分布不均的问题;最后使用1×1的卷积层将特征图回归为密度图。实验结果表明,与对比算法相比,所提算法的平均绝对误差(mean absolute error,MAE)和均方误差(mean square error,MSE)均显著降低,可取得较好的人群计数效果。

In order to solve the problem of difficulty in counting due to the large crowd density and uneven distribution,an algorithm based on scale space pyramid network was proposed.Firstly,the first ten layers of classic network VGG16 were migrated as front-end modules to extract features.Then,the scale space pyramid module designed was used to extract the multi-scale crowd features,and the dilated convolution was introduced in this module to further optimize the feature map by increasing the receptive field,so as to solve the problem of crowd crowding and uneven distribution of the crowd.Finally,the convolution layer of 1×1 was used to regress the characteristic graph to density graph.The experimental results show that the mean absolute error(MAE)and mean square error(MSE)of this algorithm are significantly reduced compared with the contrastalgorithm,and a better counting effect is achieved.