Abstract
Digital twins are virtual 3D models of real physical objects. Use cases in industry are documentation, inspection, predictive maintenance, and simulation of processes. Other fields like virtual/augmented reality, cultural heritage, or e-commerce also have a desire for realistic 3D models. Objects with unique characteristics and with unknown base data, often originating from design, are digitized by a 3D sensor. Multi-view photogrammetry (MVP) is a common 3D digitization method because it merely requires a single camera to record an object. MVP captures the 3D shape and color texture simultaneously, which allows more comprehensive digital twins. However, MVP also has the limitation that surfaces must contain sufficient texture features. Otherwise, the 3D digitization can create artifacts on the 3D models or fail entirely. Surface preparation, e.g. with stick-on markers, is a common method to counteract this issue. We present a method which utilizes virtual texture features by an object-static pattern projector. Our innovation lies in the usage of non-visible patterns in near-infrared (NIR) spectrum above 780 nm to avoid interference with the true color texture. The recording sensor is extended to a multimodal setup, consisting of one color camera and one NIR camera. Synchronous image capture enables the application of our method in handheld sensor units. Color and pattern images are combined in a customized photogrammetric workflow to reconstruct more realistic 3D models with true color texture. An accuracy better than 0.2 mm was achieved. Furthermore, we analyzed some aspects of our method with regard to its introduction to real-world applications.