基于点对点网络的人群计数方法  

Crowd Counting Method Based on Deep Learning Algorithm

作　　者：肖克爽 康志煊 XIAO Keshuang;KANG Zhixuan(Hubei University of technology,Wuhan Hubei 430070,China)

机构地区：[1]湖北工业大学,湖北武汉430070

出　　处：《信息与电脑》2022年第13期79-85,共7页Information & Computer

摘　　要：随着国内社会的快速发展,大量人口聚集的生活工作的场所隐藏着诸多的安全隐患。因此,视频监控等图像技术手段成为了维护社会治安的重要手段,人群计数作为其中一项重要的技术也得到飞速发展,在公共安全、疫情管控、应急指挥等方面得到广泛应用。近些年,随着卷积神经网络技术的发展,基于深度学习的人群计数的精确度得到了极大的提升,但由于人群遮挡、目标尺度变化、目标分布不均匀等问题,人群计数仍有提升空间。为了进一步提升计数算法的性能,本文制作了人群密度图数据集Hbut,基于pytorch框架,分别在ShanghaiTech、Hbut数据集上对点对点网络进行测试,并使用评价指标对计数性能进行评估。经过在不同数据集的实验结果和相关对比发现,证明基于点对点的人群计数算法具有很高的计数精确度。With the rapid development of the domestic society,a large number of people gathered in the places of life and work hidden many security risks.Therefore,video surveillance and other image technology tools have become an important means to maintain social security,and crowd counting as one of the important technologies has also been rapidly developed and widely used in public security,epidemic control,emergency command and other aspects.In recent years,with the development of convolutional neural network technology,the accuracy of crowd counting based on deep learning has been greatly improved,but there is still room for improvement in crowd counting due to crowd occlusion,target scale change,uneven target distribution and other problems.In order to further improve the performance of the counting algorithm,this paper produces the crowd density map dataset Hbut,tests the point-to-point network on ShanghaiTech and Hbut datasets respectively based on the pytorch framework,and evaluates the counting performance using evaluation metrics.After the experimental results and related comparisons in different datasets,it is found that the point-to-point based crowd counting algorithm is proved to have high counting accuracy.

关 键 词：人群计数 深度学习 点对点网络 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]