3D Flame Reconstruction via Improved NeRF and Radiative Transfer Modeling

Confronted with the constraints of traditional 3D tomographic technology in the reconstruction of flame fields, we introduce a 3D flame reconstruction method grounded in deep learning. Traditional techniques face challenges in high-resolution imaging, camera array configuration, and complex lighting conditions, which restrict their potential in industrial applications. To overcome these shortcomings, this paper introduces Neural Radiance Fields (NeRF) technology, an innovative implicit representation method capable of generating high-resolution, realistic flame images. The primary contribution of this paper is the combination of NeRF technology with a radiative transfer model, deriving cumulative transmittance and body reconstruction formulas, which enhance the precision of flame radiative characteristic simulation. Through the deep learning model, NeRF, augmented by the radiative transfer model, is capable of implicitly learning and representing the absorption coefficient and color information of the flame, enabling the rendering of high-fidelity images from any perspective. The model constructed in this paper not only improves the accuracy of flame 3D reconstruction but also reduces the dependence on camera calibration, saving system costs. Experimental results show that the method proposed in this paper can effectively reconstruct high-resolution instantaneous flame morphologies from sparse viewpoints, providing a new perspective for the study of 3D flame field information and contributing to the development of more efficient and environmentally friendly industrial combustion processes.