轻量级城管事件智能检测系统设计  

The design of lightweightintelligent detection system for urban management events

作　　者：唐颖 罗胜强 杨利平[2] 刘建伟 张林鹏 毛莉娟 田海波 李正浩 TANG Ying;LUO Shengqiang;YANG Liping;LIU Jianwei;ZHANG Lipeng;MAO Lijuan;TIAN Haibo;LI Zhenghao(Chongqing Info Pro Technology Co.,Ltd,Chongqing 404100 China;Chongqing University,Chongqing 400044 China;Chongqing Institute of Green and Intelligent Technology,Chinese Academy of Sciences,Chongqing 400714 China;Urban Management And Law Enforcement Detachment of Banan District Chongqing,401320 China)

机构地区：[1]重庆因普乐科技有限公司,重庆404100 [2]重庆大学,重庆400044 [3]中国科学院重庆绿色智能技术研究院,重庆400714 [4]重庆市巴南区城市管理综合行政执法支队,重庆401320

出　　处：《自动化与仪器仪表》2021年第9期165-168,共4页Automation & Instrumentation

基　　金：重庆市技术创新与应用发展专项面上项目(No.cstc2019jscx-msxmX0299);重庆市渝北区科技计划项目(No.2020-03)。

摘　　要：智能化和精细化是我国城市管理工作的发展方向。融合应用基于快速区域卷积神经网络的事件检测分类算法和基于OpenVINO的网络优化加速方法,结合网格化城市管理新趋势,提出了适合村镇、街道、园区级别管理主体,集实时性强、准确率高、全时段监控等优点于一身的轻量级城管事件智能检测系统设计方案。采用真实城管监控视频进行测试,城管事件平均分类准确率为96.14%,交并比为91.59%,可助力城市管理执法质量和效率的双重提升。Intellectualization and refinement are the trend of urban management in China.Combining the object detection and classification algorithm based on Faster R-CNN and the optimization method based on OpenVINO,in conjunction with the new trend of grid management,a scheme of a lightweight intelligent detection system for urban management event is proposed,which is suitable for village,street and park level management subjects.It integrates the advantages of strong real-time performance,high accuracy and full-time monitoring.Using urban management surveillance video for testing,the average classification accuracy of urban management events is 96.14%,and the ratio of Intersection over Union is 91.59%,which can help to improve the quality and efficiency of urban management.

关 键 词：城市管理 智能检测 快速区域卷积神经网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]