Over the past few years, human 3D reconstruction technology has seen rapid developments, enabling us to capture the shape and appearance of a person, and to automatically generate a static 3D model that looks similar to that person. In order to make this 3D model move, either through animations or motion capture, it is necessary to create a skeleton for the 3D model, which works in the same way as the human skeleton. This process is called rigging. Traditionally, this rigging process is done by an expert user manually, which can take a lot of effort and time. Recently, various rigging systems have been developed in order to simplify this process. After obtaining the 3D model of a person and rigging it, we get a animateable 3D model that looks like that person, also known as an avatar for that person.
With the abundance of both human 3D reconstruction systems and rigging systems, it can be difficult to determine which reconstruction system works well with which rigging system in order to generate a realistic avatar of a person in terms of appearance and movement. In this study, we aim to help answer this question by comparing avatars generated by different combinations of human 3D reconstruction and rigging systems. Among the reconstruction methods explored (itSeez3d, Autodesk Recap Pro, COLMAP, COLMAP with OpenMVS), COLMAP yielded the best-looking avatar, while among the rigging systems (USC autorigger, Kinect, OpenPose, Autodesk Maya autorigger), the USC autorigger yielded the best performance in terms of motion appearance. Finally, the combination of COLMAP and the USC autorigger yielded the best avatar in terms of appearance and rigging quality.