Fast capture of textured full-body avatar with RGB-D cameras

• DOI: 10.1007/s00371-016-1245-9
• Publication Date: 2016-05-06T08:29:54Z
• Authors: Shuai Lin; Yin Chen; Yu-Kun Lai; Ralph R. Martin; Zhi-Quan Cheng
• Keywords: RGB color model; Image stitching

We present a practical system which can provide\ud a textured full-body avatar within three seconds. It uses sixteen\ud RGB-depth (RGB-D) cameras, ten of which are arranged\ud to capture the body, while six target the important\ud head region. The configuration of the multiple cameras is\ud formulated as a constraint-based minimum set space-covering\ud problem, which is approximately solved by a heuristic algorithm.\ud The camera layout determined can cover the fullbody\ud surface of an adult, with geometric errors of less than\ud 5 mm. After arranging the cameras, they are calibrated using\ud a mannequin before scanning real humans. The 16 RGB-D\ud images are all captured within 1 s, which both avoids the\ud need for the subject to attempt to remain still for an uncomfortable\ud period, and helps to keep pose changes between different\ud cameras small. All scans are combined and processed\ud to reconstruct the photo-realistic textured mesh in 2 s. During\ud both system calibration and working capture of a real\ud subject, the high-quality RGB information is exploited to assist\ud geometric reconstruction and texture stitching optimization.