Multimodal multiview photogrammetry system for on-site 3D digitization of low-textured objects

Traditional multi-view photogrammetry (MVP) faces significant challenges in reconstructing low-textured objects, often resulting in artifacts or incomplete 3D models. Our work introduces a novel multimodal MVP approach that combines near-infrared (NIR) pattern projection with synchronized RGB-NIR imaging to overcome these limitations. By projecting dense, invisible NIR patterns (830. 850 nm) onto low-textured surfaces, our method enables robust feature detection while preserving the object's original color texture. A handheld dual-camera rig - comprising an RGB camera and an NIR-sensitive camera - captures both modalities synchronously, allowing dynamic, real-world scanning without surface preparation. Key innovations include a pre-calibrated camera setup for pose alignment and a customized photogrammetric workflow that fuses NIR-derived geometry with RGB texture data. The use of compact diffractive optical element (DOE) NIR projectors ensures scalability for diverse applications, from e-commerce to predictive maintenance. Experiments demonstrate that our method reduces artifacts by up to 60 times in tie point density compared to classic RGB MVP, achieving sub-0.2 mm accuracy against structured-light reference models. Practical validation on industrial and cultural heritage objects (e.g., a high-pressure turbine) confirms our system's ability to digitize complex, low-textured surfaces while maintaining portability and cost-efficiency. Our work advances handheld, dynamic MVP systems by addressing critical gaps in handling uncooperative surfaces, offering a non-invasive solution for high-fidelity digital twins. Our results underscore the potential of multimodal sensing to enhance 3D reconstruction accuracy and realism, paving the way for broader adoption in industrial and heritage preservation contexts.