基于人体关键点的着装规范检测  

Dress Code Detection Based on Human Key Points

作　　者：刘思琨 曹强 张俊杰[1] 陈健[1] 

机构地区：[1]上海大学通信与信息工程学院,上海200444

出　　处：《工业控制计算机》2023年第9期25-27,共3页Industrial Control Computer

基　　金：上海市科委项目(22511100902)。

摘　　要：为了对变电站现场人员工作服着装规范性进行有效识别,提出了一种基于人体关键点的着装规范检测算法。使用YOLOv5与SimplePose来提取人体关键点,并在HSV色彩空间中得到肤色分割信息,对人体四肢进行间隔采样,并联合肤色信息进行判断,最终对手部区域进行定位并消除因手部区域带来的误检。实验结果表明该方法具有较高精度,在自建数据集上达到94%的准确率,F1值为0.94,相较于现有方法,其不依赖数据集并且可进行更加细粒度的着装规范检测。In order to effectively identify the dress code of substation field personnel,a dress code detection algorithm based on human key points is proposed.Use YOLOv5 and SimplePose to extract human key points,and get skin color segmentation information in HSV color space.Perform interval sampling on human limbs and judge with skin color information.Finally,locate the hand area and remove the false detection caused by the hand area.The experimental results show that the method has a high accuracy,reaching 94%accuracy on the self built dataset,and the F1 value is 0.94.Compared with the existing methods,the method does not rely on the dataset and can perform more fine-grained dress code detection.

关 键 词：着装规范 人体关键点 肤色分割 HSV YOLOv5 SimplePose 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TM63[自动化与计算机技术—计算机科学与技术]