Thermal infrared scene reconstruction from UAV imagery is challenging due to low spatial resolution, view dependent radiative effects, and temporal temperature variation. Existing 3D reconstruction methods largely treat thermal measurements as static appearance signals, limiting their ability to produce stable and physically consistent reconstructions. We propose a physics guided Gaussian splatting framework for thermal scene reconstruction that incorporates models of atmospheric attenuation and heat transfer to better constrain thermal radiance estimation. By integrating physical priors into the rendering and optimization process, our method reduces floating artifacts and improves edge stability compared to vanilla 3D Gaussian Splatting. Experiments on UAV thermal datasets demonstrate more accurate and perceptually consistent reconstructions, motivating future extensions toward dynamic thermal modeling.<\/p>\n","protected":false},"excerpt":{"rendered":"
Thermal infrared scene reconstruction from UAV imagery is challenging due to low spatial resolution, view dependent radiative effects, and temporal temperature variation. Existing 3D reconstruction methods largely treat thermal measurements as static appearance signals, limiting their ability to produce stable and physically consistent reconstructions. We propose a physics guided Gaussian splatting framework for thermal scene … <\/p>\n