基于改进的Faster-RCNN的人群密度预警方法  

Improved Faster-RCNN-based Crowd Density Warning Method

作　　者：常珍 CHANG Zhen(Taiyuan Emergency Management Comprehensive Administrative Law Enforcement Detachment,Taiyuan Shanxi 030001)

机构地区：[1]太原市应急管理综合行政执法支队,山西太原030001

出　　处：《软件》2023年第10期86-88,共3页Software

摘　　要：本文提出了一种改进的基于Faster-RCNN的人群密度预警方法。通过引入软非极大值抑制算法,对Faster-RCNN进行了优化,显著提升了对密集人群的检测能力。经过改进的算法在测试集上的平均绝对误差和均方误差分别降低至3.4和9.8,表现出色。该方法不仅可用于行人检测,还能实时生成人群密度热力图,并根据平均密度分级划分拥挤程度。This paper presents an enhanced crowd density warning method based on Faster-RCNN. By incorporating the Soft Non-Maximum Suppression algorithm, Faster-RCNN has been optimized to significantly improve its detection capability for dense crowds. The improved algorithm achieved impressive results, reducing the mean absolute error and mean square error on the test dataset to 3.4 and 9.8 respectively. This approach is not only applicable for pedestrian detection, but it also has the capability to generate real-time crowd density heatmaps and classify congestion levels based on average density.

关 键 词：公共安全 人群密度估计 机器视觉 Soft-NMS算法 Faster-RCNN算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]