3:30 PM - 5:00 PM
[SVC35-P15] Development of Technique for Grasping Volcano Disasters in Real Time Using Unmanned Aerial Vehicles (UAVs): Current status and future directions
Keywords:UAV, Volcano Disaster, Real Time, 3D model
Since it is dangerous for manned aircraft to fly over volcanic craters during volcanic eruptions, unmanned aerial vehicle (UAV) surveys are expected to be used for this purpose. The Ministry of Education, Culture, Sports, Science and Technology's Integrated Program for Next Generation Volcano Research and Human Resource Development has been studying " Development of Technique for Grasping Volcano Disasters in Real Time Using Unmanned Aerial Vehicles (UAVs)" since 2016, since 2016, and this paper will present the current status and future prospects of the research.
The prerequisites for conducting this research include the use of commercially available UAVs (multi-rotor type), the use of visible and thermal images, and the absence of lasers or other sensors. The basic analysis is to create a three-dimensional model using SfM-MVS (Structure from Motion / Multi-view Stereo) technology. Point cloud data, Digital Surface Models (DSM), and orthoimages can be created, but in some cases, data similar to Digital Elevation Models (DEM) can be created, as there may be little vegetation around the crater area. Red Relief Image Map and elevation maps can be created from the three-dimensional models, which can be used for terrain interpretation and other purposes.
The major challenge in the first half of the research was to improve the accuracy of the 3D model. Aso volcano erupted in 2016, when this study was started, and vertical photographs were taken around the Nakadake crater using a UAV. Because of the limited time available within the restricted entry zone, no Ground Control Point (GCP) was established and existing structures were used. We attempted to perform a difference analysis with the existing aerial lidar DEM, but the lack of a GCP prevented good positional adjustment, so a highly accurate difference analysis could not be performed. Subsequent demonstration tests on Izu Oshima volcano confirmed that the accuracy of the 3D model improved when GCPs were installed, but it is expected to be difficult to install GCPs near objects during volcanic eruptions. Therefore, we verified that UAVs equipped with RTK (Real-Time Kinematic) from 2020 can be used to create 3D models at the same level as when GCPs are installed, even in the absence of GCPs. The use of RTK-equipped UAVs also reduced the time required to create the 3D models.
The 3D model is expected to be used to estimate the thickness of ejecta by difference analysis with existing data, and to determine the distribution of craters and ejecta by using a Red Relief Image Map. In the demonstration experiment on Izu Oshima volcano, a detailed three-dimensional model of the central crater of Mt. Mihara was created from a large number of photographs of the vertical crater walls, which are difficult to measure with aerial lidar surveying. The detailed topographic model revealed the volume of the crater and the altitude at which lava overflows from the crater. Therefore, if the elevation of the lava surface can be determined in real time using an oblique photogrammetry system by taking oblique photographs of the crater from a distance using a UAV or other equipment, it will be possible to predict the timing of lava overflow.
On the other hand, we are also studying a method that does not create a three-dimensional model, but uses a method used in the infrastructure field to automatically patrol predetermined points, take visible and thermal images, and extract changes. Since thermal images can be used to monitor conditions even at night, there is a possibility that the formation of craters, outflow of lava, etc., and expansion of geothermal zones can be monitored in real time. In addition, since the current eruption warning level (JMA) is based on the distance of ballistic projectiles, thermal images taken by UAVs are expected to be used to quickly determine the location of ballistic projectiles when they are scattered.
We think that we are almost at the end of the road to creating highly accurate three-dimensional models from images taken by UAVs, but there is still room for improvement in speeding up data processing. In addition, little research has been done on how to automatically extract the distribution range of ejecta from the obtained 3D models and images using AI and machine learning. In the future, we would like to develop technologies to grasp the eruption status of active volcanoes in real time from images acquired by UAVs using automatic extraction and other methods, and to provide this information quickly from the volcanic field.
The prerequisites for conducting this research include the use of commercially available UAVs (multi-rotor type), the use of visible and thermal images, and the absence of lasers or other sensors. The basic analysis is to create a three-dimensional model using SfM-MVS (Structure from Motion / Multi-view Stereo) technology. Point cloud data, Digital Surface Models (DSM), and orthoimages can be created, but in some cases, data similar to Digital Elevation Models (DEM) can be created, as there may be little vegetation around the crater area. Red Relief Image Map and elevation maps can be created from the three-dimensional models, which can be used for terrain interpretation and other purposes.
The major challenge in the first half of the research was to improve the accuracy of the 3D model. Aso volcano erupted in 2016, when this study was started, and vertical photographs were taken around the Nakadake crater using a UAV. Because of the limited time available within the restricted entry zone, no Ground Control Point (GCP) was established and existing structures were used. We attempted to perform a difference analysis with the existing aerial lidar DEM, but the lack of a GCP prevented good positional adjustment, so a highly accurate difference analysis could not be performed. Subsequent demonstration tests on Izu Oshima volcano confirmed that the accuracy of the 3D model improved when GCPs were installed, but it is expected to be difficult to install GCPs near objects during volcanic eruptions. Therefore, we verified that UAVs equipped with RTK (Real-Time Kinematic) from 2020 can be used to create 3D models at the same level as when GCPs are installed, even in the absence of GCPs. The use of RTK-equipped UAVs also reduced the time required to create the 3D models.
The 3D model is expected to be used to estimate the thickness of ejecta by difference analysis with existing data, and to determine the distribution of craters and ejecta by using a Red Relief Image Map. In the demonstration experiment on Izu Oshima volcano, a detailed three-dimensional model of the central crater of Mt. Mihara was created from a large number of photographs of the vertical crater walls, which are difficult to measure with aerial lidar surveying. The detailed topographic model revealed the volume of the crater and the altitude at which lava overflows from the crater. Therefore, if the elevation of the lava surface can be determined in real time using an oblique photogrammetry system by taking oblique photographs of the crater from a distance using a UAV or other equipment, it will be possible to predict the timing of lava overflow.
On the other hand, we are also studying a method that does not create a three-dimensional model, but uses a method used in the infrastructure field to automatically patrol predetermined points, take visible and thermal images, and extract changes. Since thermal images can be used to monitor conditions even at night, there is a possibility that the formation of craters, outflow of lava, etc., and expansion of geothermal zones can be monitored in real time. In addition, since the current eruption warning level (JMA) is based on the distance of ballistic projectiles, thermal images taken by UAVs are expected to be used to quickly determine the location of ballistic projectiles when they are scattered.
We think that we are almost at the end of the road to creating highly accurate three-dimensional models from images taken by UAVs, but there is still room for improvement in speeding up data processing. In addition, little research has been done on how to automatically extract the distribution range of ejecta from the obtained 3D models and images using AI and machine learning. In the future, we would like to develop technologies to grasp the eruption status of active volcanoes in real time from images acquired by UAVs using automatic extraction and other methods, and to provide this information quickly from the volcanic field.