基于改进CSRNet的人群计数算法

Crowd counting algorithm based on improved CSRNet

为实现对密集人群快速和准确地计数,提出了一种基于改进CSRNet的人群计数算法。为加快训练速度,提高计数精度,针对CSRNet网络进行改进,将后端膨胀卷积网络替换为普通卷积结构,同时增加了密集连接结构。改进后的算法分别在ShanghaiTech A,B数据集和UCF_CC_50数据集上进行训练和测试,实验结果表明,相比其他人群计数算法,改进后的算法具有更高的计数精度。

In order to count the crowd quickly and accurately,a crowd counting algorithm based on improved CSRNet is proposed.In order to speed up the training and improve the count accuracy,CSRNet network is improved by replacing the back-end dilated convolution network with the ordinary convolution structure,and the structure of dense connection is added.The improved algorithm is trained and tested on ShanghaiTech A,B and UCF_CC_50 datasets,respectively.The experimental results show that the improved algorithm has better recognition accuracy compared to other crowd counting algorithms.