基于图像识别技术的卷烟零售数据采集方法  

Cigarette retail sales data collection method based on image recognition technology

作　　者：梁冬[1] 陈智斌[1] 农英雄[1] 潘剑 汪倍贝[1] LIANG Dong;CHEN Zhibin;NONG Yingxiong;PAN Jian;WANG Beibei(China Tobacco Guangxi Industrial CO.,LTD.,Nanning 530001,China)

机构地区：[1]广西中烟工业有限责任公司,南宁市530001

出　　处：《中国烟草学报》2021年第5期81-89,共9页Acta Tabacaria Sinica

基　　金：广西中烟工业有限责任公司科技计划项目“基于图像识别的卷烟零售数据采集与分析技术研究”(GXZYCX2019E002)。

摘　　要：为了实现卷烟零售数据的自动化采集,提出了一种基于图像识别技术的卷烟零售数据采集方法。首先,基于深度图设计一种人体检测与跟踪算法,稳定跟踪消费者在购买卷烟期间的行动轨迹;其次,实现一种7点式的人体关键点检测算法,检测顶式视角下的人体关键点;然后,构建卷烟的检测与识别算法,检测卷烟在图像中位置与姿态并识别卷烟的牌号;最后,融合人体跟踪信息、人体关键点信息、卷烟位置和牌号信息构建一种消费者行为识别方法,实现卷烟零售数据的采集。结果表明:数据采集方法的准确率达到95.6%,通过图像识别技术实现卷烟零售数据的自动化采集是可行的。In order to realize automatic collection of cigarette retail sales data,a method based on image recognition technology is proposed in this study.Firstly,a human detection and tracking algorithm based on depth image was designed to stably track the consumers'movement trajectory during the purchase of cigarettes.Secondly,a 7-point human keypoint detection algorithm was implemented to detect the human key points from the top view.Then,a cigarette detection and recognition algorithm was constructed to detect the cigarettes’position and pose,and then recognize the brand of cigarettes.Finally,a consumer behavior recognition method was constructed by integrating human tracking information,human keypoint information,cigarette location and brand information to realize the cigarette retail sales data collection.The results show that the accuracy of the proposed data collection method is 95.6%,which demonstrates that it is feasible realize the automatic collection of cigarette retail sales data based on image recognition technology.

关 键 词：卷烟零售数据 图像识别 人体跟踪 人体关键点 卷烟牌号识别 行为识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] F721[自动化与计算机技术—计算机科学与技术]