{"title":"A New Bottom-up Human Pose Estimation Method by Body Center and Anchor Points","authors":"Jiahua Wu, H. Lee","doi":"10.1109/ICCEAI52939.2021.00047","DOIUrl":null,"url":null,"abstract":"There are two stages in bottom-up human pose estimation method, joint detection and joint candidate grouping. Optimizing grouping algorithms can significantly improve the performance of pose estimation. In this paper, we introduce body center, a center point of person instance, and anchor point, a corresponding assistant position point, for the task of grouping. The anchor point is the center of joint and body center, which can help joint grouping to the corresponding person instance like an anchor. The body center and anchor point can be predicted simultaneously with the joint candidate by the same backbone. So, this new grouping method can fully exploit the features extracted by the step of joint detection. On the COCO keypoints dataset, the proposed method performs on par with the existing state-of-the-art bottom-up method in accuracy.","PeriodicalId":331409,"journal":{"name":"2021 International Conference on Computer Engineering and Artificial Intelligence (ICCEAI)","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2021-08-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2021 International Conference on Computer Engineering and Artificial Intelligence (ICCEAI)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/ICCEAI52939.2021.00047","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
There are two stages in bottom-up human pose estimation method, joint detection and joint candidate grouping. Optimizing grouping algorithms can significantly improve the performance of pose estimation. In this paper, we introduce body center, a center point of person instance, and anchor point, a corresponding assistant position point, for the task of grouping. The anchor point is the center of joint and body center, which can help joint grouping to the corresponding person instance like an anchor. The body center and anchor point can be predicted simultaneously with the joint candidate by the same backbone. So, this new grouping method can fully exploit the features extracted by the step of joint detection. On the COCO keypoints dataset, the proposed method performs on par with the existing state-of-the-art bottom-up method in accuracy.