密集场景下的人流密度统计方法研究综述  

An Overview of Statistics Methods of People Flow Density in Dense Scenes

作　　者：郭淑涛 韩琳 GUO Shu-Tao;HAN Lin(The First Research Institute of the Ministry of Public Security of P.R.C,Beijing100044,China)

机构地区：[1]公安部第一研究所,北京100044

出　　处：《机电产品开发与创新》2022年第6期162-165,169,共5页Development & Innovation of Machinery & Electrical Products

摘　　要：密集场景下的人流密度统计方法的研究是一项近年来出现的计算机视觉领域的具有挑战性的难题,密集人群有人数多,人群分布情况复杂的特点。该课题结合深度学习技术后,取得了一定程度的研究进展。当前该领域研究方法可以归纳为两大类:一类是基于回归的人数统计方法研究,另一类是基于检测的人数统计方法研究。本文对两大类方法分别进行了系统的归纳总结,整理了目前常用的数据集和对应的评价指标,并比较各个数据集之间的特点,总结该课题当前存在的问题和研究难点,展望未来的研究趋势。The study of crowd density statistical method in dense scenes is a challenging problem in the field of computer vision emerging in recent years,the dense population is characterized by a large number of people and a complex population distribution.After combining with deep learning technology,this topic has made a certain degree of research progress.The current research methods in this field can be summarized into two categories:one is the regression-based numerical statistics method,and the other is that based on detection.This paper systematically summarizes the two kinds of methods,sorts out the commonly used data sets and corresponding evaluation indicators,compares the characteristics of each data set,summarizes the current problems and research difficulties of the subject,and looks forward to the future research trend.

关 键 词：深度学习 密集场景 人流密度 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]