摘要
基于视频分析,对实时获取车站的客流量的方法进行了研究。对获得的视频图像进行了背景分离以及补偿填充与合并操作,设计了一种改进的Hough变换算法,并引入重合圆判断条件,以此提取人体头部区域的类圆形特征,采用基于区域的跟踪方法以及计数线的方法实现目标的跟踪计数。通过实验分析得到,采用所提计数方法,识别出每帧图像的平均时间约为193 ms,识别的准确率达到了92.33%,造成误差的主要原因为误检外物类圆特征,其误差率约为6.67%,较为符合实际车站客流计数的应用需求。
In order to obtain the passenger flow at the station in real time,this paper studies it based on video analysis.The background separation and compensated filling and merging of the obtained video images are done.An improved Hough transform algorithm is designed,and the coincident circle criterion is introduced to extract the circular-like features of human head area.The area-based tracking method and the counting line method are used to achieve the target tracking count.Through experimental analysis,using the counting method in this article,the average time to recognize each frame of image is about 193 ms,and the recognition accuracy rate reaches 92.33%.The main cause of the error is the misdetection of the foreign object circle feature,and the error rate It is about 6.67%,which is more in line with the application requirements of actual station passenger flow counting.
作者
陶征勇
宗起振
李佑文
肖振远
党聪
TAO Zhengyong;ZONG Qizhen;LI Youwen;XIAO Zhenyuan;DANG Cong(Nanjing Guodian Nanzi Rail Transit Engineering Co.,Ltd.,Nanjing 210000,China)
出处
《微型电脑应用》
2023年第5期103-106,128,共5页
Microcomputer Applications
关键词
客流计数
背景分离
补偿填充与合并
改进的Hough变换
重合圆
passenger flow counting
background separation
compensation filling and merging
improved Hough transform
coincidence circle
作者简介
陶征勇(1994-),硕士,助理工程师,研究方向为图像识别、智能检测与控制;宗起振(1976-),硕士,高级工程师,研究方向为电力设备在线监测及故障检修领域;李佑文(1985-),硕士,高级工程师,研究方向为数据挖掘技术、智慧城轨、智慧车站;肖振远(1994-),硕士,助理工程师,研究方向为机器视觉;党聪(1993-),硕士,助理工程师,研究方向为图像识别。
