[S-VC34] Monitoring and assessment of volcanic activities

[SVC34-P13] Direct georeferencing approach to reconstructing 3D thermal photogrammetric models of active volcanoes

*Nashimoto Subaru¹, Yokoo Akihiko¹ (1.Graduate School of Science, Kyoto University)

Keywords:Thermal photogrammetric imaging, SfM-MVS, direct georeferencing

Thermal photogrammetric imaging has been recently proposed as a new technique for monitoring active volcanoes (Thiele et al., 2017). It generates 3D thermal photogrammetric models through Structure from Motion and Multi-View Stereo (SfM-MVS) process with visible and thermal infrared (TIR) images. This technique has been extensively studied in Photogrammetry, mainly targeting industrial environments. Many studies generate 3D point clouds with TIR images and ones with visible images separately and align them through point cloud registration algorithms (Lopez et al., 2021). However, this approach has some drawbacks. SfM processes with TIR images usually result in sparser and less accurate 3D point clouds than the ones generated with visible images due to low resolution and aberration-induced blurring of TIR images (Lopez et al., 2021). This approach also does not allow users to correct the effect of atmospheric attenuation of the infrared wave, which results in underestimating the temperature of the target objects. In order to solve these problems, we propose the direct georeferencing strategy (Turner et al, 2013). We used visible and TIR image datasets acquired with a TIR camera (FLIR Zenmuse XT2) deployed on an unmanned aerial vehicle (DJI Matrice 200). With this camera, we can take visible and TIR images simultaneously. We first generate a Digital Elevation Model (DEM) through the SfM-MVS process with visible images, using commercial software (Agisoft Metashape Professional). In this process, we can also get the positions and orientations of the camera. Then, with these parameters and a camera model, we calculate a bearing vector for each pixel of the TIR images in the camera reference frame, which points from the pixel to the corresponding target on the DEM. We extend the vector until it intersects the DEM surface and obtain the distance between the pixel and the target, which enables pixel-wise correction of the atmospheric attenuation. Then, we assign the corrected temperature to the DEM coordinate of the intersection (direct georeferencing). By georeferencing every pixel of all the TIR images, we reconstructed a 3D thermal photogrammetric model. We applied our method to data of the Nakadake 1st crater of Aso volcano. We generated a 1-m resolution DEM from 755 visible images. Then, we georeferenced 755 TIR images and reconstructed a 3D thermal photogrammetric model. We tested the accuracy of our method. First, we manually selected control points and read their coordinates both on the TIR images and the DEM. Then, we georeferenced the control points on the TIR images and compared their coordinates with the ones we read on the DEM. As a result, the error was a few meters. We regarded this error as acceptable, considering the mesh size of the DEM (1-m) and the errors resulting from manually reading the coordinates of the control points on the TIR images. To highlight the usefulness of this model, we calculated the heat discharge rate from the fumarolic area on the south crater wall of the crater, based on the heat balance method proposed by Sekioka and Yuhara (1974). Our result was 1.4 MW and consistent with the value, 1.1 MW, calculated from a single TIR image assuming constant line-of-sight distance (Yokoo and Ishii, 2021).