A research team at Carnegie Mellon University in Pennsylvania is developing a technology to create a 3D model by scanning a human face by simply taking a slow-motion video on a smartphone. This technology uses an AI algorithm, and the slow motion to become a material ends in 15-20 seconds. The 3D model is created in about 30 minutes.
It is not easy to reproduce the human face. Something subtly different due to this problem may damage reality. However, the research team explains that it employs a two-step process to create an accurate 3D model.
First, the relative distance and location of the smartphone and the subject are calculated by triangulation, which simultaneously estimates self-location and creates environmental maps called SLAM (Simultaneous Localization and Mapping). Here, an algorithm based on deep learning detects feature features with an accuracy of less than 1mm. At this time, the accuracy of the 3D model is improved with a computer vision technology called mesh fitting. Related information can be found here .