A 3D Gaussian Splatting method from Tohoku University and Sendai startup XMAT has been registered in NETIS, the new technology database run by Japan's infrastructure ministry.
Tohoku University says it is the first 3D Gaussian Splatting technology registered in NETIS anywhere in Japan. The Ministry of Land, Infrastructure, Transport and Tourism lists it under a Japanese title that translates as high definition 3D model generation by the Gaussian splatting method. The method builds models of bridges, tunnels and plants dense with piping and equipment from images shot on commercial 360 degree cameras, ordinary cameras or drones.
Three dimensional records of those structures have mostly come from laser scanning or photogrammetry, which the university notes can require expensive dedicated equipment and survey work. The registered method does not assume that equipment. To fix scale, a reference jig of known dimensions is placed in the scene during capture, and the whole 3DGS space is aligned to its measurements. Consumer camera makers have started offering splat reconstruction too, with the Insta360 X6 turning 360 captures into gaussian splats through its cloud-processed Spatial Capture feature.
The team has built a free viewer that opens the models on an ordinary laptop, without a workstation or point cloud processing software. The university attributes the light file sizes to splats rendering smoothly even at low point density. Inspection photos, damage records, nondestructive testing results and comments can be pinned to positions in the 3D space, and orthographic images can be exported from any viewpoint.
The developers are Akira Yoshikawa, Hiroki Sato and Kenji Inoue of Tohoku University's New Industry Creation Hatchery Center, working with XMAT, a Tohoku University spinout. The same team has been combining nondestructive testing with augmented reality to display measurements alongside their positions, work that led to a 2022 NETIS registration for a portable X-ray fluorescence analyzer that measures chloride in concrete. It now plans to link the 3DGS models with that AR work, using AR to capture and display information on site and 3DGS to store and share the site as a digital space.
Tohoku University has not published accuracy, capture time or processing time figures for the method.
The full details are in Tohoku University's announcement.
