Thermal Point Cloud Generation and Evaluation from Uncalibrated Unmanned Aerial Vehicle–Visible and Infrared Camera Images Using Position and Orientation System Prior

This paper presents a methodology for thermal infrared (TIR) point cloud acquisition, leveraging information from the visible camera, TIR camera, and their respective unmanned aerial vehicle (UAV) position and their position and orientation systems (POSs) prior. First, the visible images and their corresponding UAV POS information are used to generate a red–green–blue (RGB) dense point cloud with geographic coordinates; subsequently, using the RGB dense point cloud as a reference, the acquisition of a calibration parameter between global navigation satellite system/inertial measurement unit and the TIR camera is conducted using the POS information corresponding to the selected TIR images; finally, a TIR point cloud of the entire survey area is generated based on the calibration parameters. The availability and suitability of the fused TIR point cloud generated and UAV POS data are also evaluated in terms of localization deviations and orientation deviations. The experimental results demonstrate that the acquisition of coarse-quality TIR point cloud can be achieved by using a priori POS data provided by different small UAVs. The TIR point cloud generated by the UAV POS data guidance can be directly applied to medium- and large-sized buildings. The POS-guided projected images achieved an 86.67% registration success rate with the TIR images, indicating that the orientation deviations are within acceptable limits and establishing a solid foundation for subsequent fine fusion.