Pengpeng Liu, Guixuan Zhang, Hu Guan, Jie Liu, Shuwu Zhang, Zhi Zengi
{"title":"Relative Pose Estimation for RGB-D Human Input Scans via Human Completion","authors":"Pengpeng Liu, Guixuan Zhang, Hu Guan, Jie Liu, Shuwu Zhang, Zhi Zengi","doi":"10.1109/ICCST53801.2021.00105","DOIUrl":null,"url":null,"abstract":"Relative pose estimation for human scans enjoys a promising prospect. However, most existing methods mainly focus on indoor or outdoor scenes, requiring considerable overlap between the inputs. We present a technique for estimating the relative pose whatever the overlap between the human RGB- D input scans is. For non-overlapping scans, the insight is to take advantage of the underlying human geometry prior as much as possible. We utilize the implicit function model for human reconstruction, enriching abundant hidden cues for unseen regions, then we use the completed human geometry to get a stable pose estimation. Our evaluation shows that our approach outperforms considerably than standard pipelines in non-overlapping setting, without compromising performance over overlapping input scans.","PeriodicalId":222463,"journal":{"name":"2021 International Conference on Culture-oriented Science & Technology (ICCST)","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2021-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2021 International Conference on Culture-oriented Science & Technology (ICCST)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/ICCST53801.2021.00105","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
Relative pose estimation for human scans enjoys a promising prospect. However, most existing methods mainly focus on indoor or outdoor scenes, requiring considerable overlap between the inputs. We present a technique for estimating the relative pose whatever the overlap between the human RGB- D input scans is. For non-overlapping scans, the insight is to take advantage of the underlying human geometry prior as much as possible. We utilize the implicit function model for human reconstruction, enriching abundant hidden cues for unseen regions, then we use the completed human geometry to get a stable pose estimation. Our evaluation shows that our approach outperforms considerably than standard pipelines in non-overlapping setting, without compromising performance over overlapping input scans.