西安工程大学学报
西安工程大學學報
서안공정대학학보
JOURNAL OF XI'AN POLYTECHNIC UNIVERSITY
2009年
2期
423-428
,共6页
服装%立体视觉%三维图像
服裝%立體視覺%三維圖像
복장%입체시각%삼유도상
garment%stereo vision%3 D image
提出了一种新开发的用于三维服装图像生成的立体视觉系统.相比于二维图像,三维服装图像为显示服装的风格、悬垂和适体性提供更真实和有效的可视化手段.该系统由4个相同的,可组合的图像单元组成,可覆盖身体的上部、下部或全身.每个单元包含一对数码相机和一台投影仪.2个相机从不同角度获取的图像用于立体匹配;投影仪产生的图案用于在布面上制造人工纹理以助于减少立体匹配的不确定性.文中采用的2个阶段的算法能实现亚像素、高密度匹配.第一阶段进行前景(服装)分割及生成初步偏移图(disparity map).第二阶段使用局域最小二乘匹配和全局优化细化偏移图及生成三维数据.然后,用曲面造型算法将无序的数据点连接成三角网格,并对其进行细分和优化,从而形成完整光滑的服装表面.
提齣瞭一種新開髮的用于三維服裝圖像生成的立體視覺繫統.相比于二維圖像,三維服裝圖像為顯示服裝的風格、懸垂和適體性提供更真實和有效的可視化手段.該繫統由4箇相同的,可組閤的圖像單元組成,可覆蓋身體的上部、下部或全身.每箇單元包含一對數碼相機和一檯投影儀.2箇相機從不同角度穫取的圖像用于立體匹配;投影儀產生的圖案用于在佈麵上製造人工紋理以助于減少立體匹配的不確定性.文中採用的2箇階段的算法能實現亞像素、高密度匹配.第一階段進行前景(服裝)分割及生成初步偏移圖(disparity map).第二階段使用跼域最小二乘匹配和全跼優化細化偏移圖及生成三維數據.然後,用麯麵造型算法將無序的數據點連接成三角網格,併對其進行細分和優化,從而形成完整光滑的服裝錶麵.
제출료일충신개발적용우삼유복장도상생성적입체시각계통.상비우이유도상,삼유복장도상위현시복장적풍격、현수화괄체성제공경진실화유효적가시화수단.해계통유4개상동적,가조합적도상단원조성,가복개신체적상부、하부혹전신.매개단원포함일대수마상궤화일태투영의.2개상궤종불동각도획취적도상용우입체필배;투영의산생적도안용우재포면상제조인공문리이조우감소입체필배적불학정성.문중채용적2개계단적산법능실현아상소、고밀도필배.제일계단진행전경(복장)분할급생성초보편이도(disparity map).제이계단사용국역최소이승필배화전국우화세화편이도급생성삼유수거.연후,용곡면조형산법장무서적수거점련접성삼각망격,병대기진행세분화우화,종이형성완정광활적복장표면.
This paper presents a newly developed stereo vision system for generating 3D images of garments attired by a subject or on a mannequin. 3D clothing images provide more realistic and effective visualization for showing clothing's style, drape, and fit than 2D images. The system consists of four identical imaging units that can be reconfigured to cover the top, bottom or entire body. Each unit contains a pair of digital cameras and a projector. The cameras capture two images from different perspecfives, and the projector creates artificial textures" onto the clothing surface that help to reduce stereo matching ambiguity. A two-phase algorithm was developed to permit sub-pixel, dense stereo matching. In the first phase, the foreground (garment) is segmented by using a predefined virtual interface in the disparity space image, and a coarse disparity map is generated with block matching. In the second phase, local least squares matching is performed along with global optimization within a regularization framework which ensures both accuracy and reliability of 3D data generation. The unorganized data clouds are the processed by a surface modeling algorithm, which is based on a triangular mesh, surface subdivision and optimization, to reconstruct smooth and complete a clothing surface.