Thermal inspection of buildings regarding efficient energy use is an increasing need in today’s energy-demanding world. This paper proposes a framework for mapping temperature attributes from thermal point clouds onto building facades. The goal is to generate thermal textures for three-dimensional (3D) analysis. Classical texture generation methods project facade images directly onto a 3D building model. Due to the limited level of detail of these models, projection errors occur. Therefore, we use point clouds from mobile laser scanning extended by intensities extracted from thermal infrared (TIR) image sequences. We are not using 3D reconstructed point clouds because of the limited geometric density and accuracy of TIR images, which can lead to poor 3D reconstruction. We project these thermal point clouds onto facades using a mapping algorithm. The algorithm uses a nearest neighbor search to find an optimal nearest point with three different approaches: “Minimize angle to normal”, “Minimize perpendicular distance to normal”, and “Minimize only distance”. Instead of interpolation, nearest neighbor is used because it retains the original temperature values. The thermal intensities of the optimal nearest points are weighted by resolution layers and mapped to the facade. The approach “Minimize perpendicular distance to normal” yields the finest texture resolution at a reasonable processing time. The accuracy of the generated texture is evaluated based on estimating the shift of the window corner points from a ground truth texture. A performance metric root-mean-square deviation (RMSD) value that measures this shift is calculated. In terms of accuracy, the nearest neighbor method outperformed bilinear interpolation and an existing TIR image-based texturing method.

Efficient and robust 3D scene representation is crucial in autonomous driving, robotics, and related fields. While RGB images provide valuable content for 3D reconstruction, other modalities like thermal or depth can enable additional information on the environment. Lately, novel view synthesis methods like 3D Gaussian Splatting have started using multiple modalities to further boost their performance. But fusing or combining multimodal data can make the process slower and can bring in additional challenges. Therefore, our project aims to use single modality based on thermal infrared domain, by removing the reliance on visible light as much as possible. This single modality can be expected to be faster as it does not rely on multimodal data. We propose a method, Thermal-to-Depth Gaussian Splatting (TDg), that uses only thermal images and depth estimation in its architecture to derive the radiance fields. Our TDg method outperforms the MSMG (Multiple Single-Modal Gaussians) baseline in most cases on our test datasets, RGBT-Scenes and ThermalMix. On average, the rendering quality metrics such as learned perceptual image patch similarity (LPIPS), structural similarity index measure (SSIM), and peak signal-to-noise ratio (PSNR) of TDg are 1.12%, 0.034%, and 0.01% better than the baseline MSMG values. It also reduces the training time significantly, by 12 mins 47 secs (55% improvement). Overall, our method is successful in deriving these thermal radiance fields, which can ultimately have several applications, such as identifying heat sources critical in surveillance, search or rescue operations, and industrial inspections where temperature is widely used to monitor machines.

Thermal infrared (TIR) imaging provides a unique capability to reveal surface heat-transfer patterns of buildings and supports applications such as energy leakage detection, insulation inspection, and building energy monitoring. However, large-scale TIR image analysis by deep learning is still constrained by the lack of reliable annotations, as TIR images often exhibit blurred textures and weak boundaries, which makes manual labeling inconsistent and time-consuming. To address this challenge, we propose ThermalAssist, a geometry-aware and gradient-enhanced framework designed to assist thermal anomaly labeling in UAV-based TIR images. By combining sparse labeling, dense correspondence, and label propagation, the framework efficiently transfers labels across overlapping views while maintaining geometric and semantic consistency. Experiments on the TBBR (Thermal Bridges on Building Rooftops) dataset demonstrate that ThermalAssist achieves an F1-score of up to 0.87 and a mean IoU of 0.69, effectively helping reduce missing annotations and inconsistent boundaries. Compared with the tracking-based SAMURAI method, our approach shows greater robustness under low-texture and low-overlap conditions. This work establishes a foundation for a broader thermal annotation system and validates an important step toward scalable, reliable, and more intelligent labeling of thermal anomaly imagery.