一种基于YOLOv3的共享单车违规停放检测方法  

An Illegal Parking Detection Approach for Shared Bicycles Based on YOLOv3

作　　者：盛宜华 武友新[1] 姚磊岳[2] SHENG Yihua;WU Youxin;YAO Leiyue(School of Information Engineering,Nanchang University,Nanchang 330031,China;Center of Collaboration and Innovation,Jiangxi University of Technology,Nanchang 330098,China)

机构地区：[1]南昌大学信息工程学院,南昌330031 [2]江西科技学院协同创新中心,南昌330098

出　　处：《计算机工程》2019年第12期237-242,共6页Computer Engineering

基　　金：江西省科技厅科技计划专项“基于自然语音交互模式的行车安全辅助系统”(20171BBE50060);南昌市科技局科技计划项目“基于移动互联网的‘人-车’语音交互系统”(2016-ZCJHCXY-013)

摘　　要：为解决共享单车随意停放给交通管理带来的困难,提出一种基于计算机视觉的共享单车违规停放检测方法。通过多尺度检测训练以及k-means维度聚类改进YOLOv3网络,在此基础上获取共享单车在图片上的特征矩阵,根据特征矩阵计算当前场景下共享单车的运行状态并进行状态统计。在交通监控视频数据集上的测试结果表明,该方法的检测准确率达到87%以上,能够实现共享单车违规停放的有效检测并给出实时预警。To solve the transportation management difficulties caused by the random parking of shared bicycles,this paper proposes an illegal parking detection method based on computer vision for shared bicycles.The method uses multi-scale detection training and k-means dimensional clustering to improve YOLOv3 network,and accordingly obtains the feature matrix of the shared bicycle on the image.Based on the obtained feature matrix,the running state of the target shared bicycle in a certain scenario can be computed for statistics.Testing results on the dataset of transportation monitoring videos show that the detection accuracy rate of the proposed method is above 87%,which means it can effectively detect illegal parking of shared bicycles and give real-time alarms.

关 键 词：共享单车 停放检测 YOLOv3网络 计算机视觉 状态统计 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]