校园拥挤踩踏事故人工智能预警与导流系统研究  

Research on Artificial Intelligence Early Warning and Diversion System for Stampede in Schools

作　　者：徐自远 丁华平 Xu Ziyuan;Ding Huaping(Wuxi Institute of Mechatronic Engineering,Jiangsu Union Technical Institute,Wuxi 214145,China;Nanjing University(Suzhou)High and New Technology Research Institute,Suzhou 215127,China)

机构地区：[1]江苏联合职业技术学院无锡机电分院,无锡214145 [2]南京大学(苏州)高新技术研究院,苏州215127

出　　处：《信息化研究》2021年第5期65-71,共7页INFORMATIZATION RESEARCH

基　　金：苏州市重点产业技术创新项目(No.SYG201917);苏州市重点产业技术创新项目(No.SYG201916);苏州市重点产业技术创新项目(No.SYG202032)。

摘　　要：基于保障学生安全的考虑,研究分析校园学生拥挤踩踏事故的成因,开发一套完善的智能预警系统,对预控该类事故具有重要的现实意义。针对传统的人群检测技术所存在的识别速度慢、鲁棒性差等问题,本文提出使用人工智能技术实现校园环境拥挤踩踏事故预警。采用基于卷积神经网络的目标检测方法,基于开源计算机视觉库和相关人工智能技术,实现对校园监控视频进行自动特征提取,实时分析人群量、速度等参数,构建人群密度、流向、滞留量、混乱程度等预警指标,搭建拥挤踩踏事故预警系统。经实验测试,所提方法能够获取实时信息,并通过显示屏与广播发布导流信息,降低踩踏事件发生的几率。Based on the consideration of ensuring student safety, researching and analyzing the causes of crowding and trampling accidents and developing a complete intelligent pre-warning system for the students have crucial practical significance in preventing such accidents. In view of the slow recognition speed, poor robustness and etc. in traditional crowd detection technique, this research utilizes novel artificial intelligence(AI) techniques to realize the pre-warning of crowded trampling accidents in the campus environment. Technically, to realize automatic feature extraction of campus surveillance video, the target detection method based on convolutional neural network, open-source computer vision libraries and related AI techniques are adopted. Analyzing parameters such as crowd volume and speed in real time and constructing pre-warning indicators including crowd density, flow direction, detention, degree of confusion and etc. to establish a pre-warning system for crowded stampede accidents. Experimental results show that the proposed method can obtain real-time information and release diversion information through display screen and broadcast, so as to reduce the probability of stampede events.

关 键 词：校园踩踏 智能预警 目标检测 自动特征提取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]