New method for creating large 3D models of urban areas is faster and cheaper

Summary

Ground truth, generated images, and visualization of Gaussian means of our Waterloo scene at different altitudes and orientations. (Left) Waterloo scene ground truth. (Middle) Waterloo scene 3DGS generated image. (Right) Waterloo scene visualization of the location of each 3DGS Gaussian, i.e., 3D positional mean of each Gaussian. These points were then extracted as point clouds. Credit: Kyle Gao et al, Enhanced 3-D Urban Scene Reconstruction and Point Cloud Densification Using Gaussian Splatting and Google Earth Imagery (2025) A research team led by Waterloo Engineering has developed a faster, cheaper way to create large-scale, three-dimensional (3D) computer models of urban areas, technology that could impact fields including urban planning, architectural design and filmmaking. A paper on the research, titled "Enhanced 3D Scene Reconstruction and Point Cloud Densification using Gaussian Splatting and Google Earth Imagery," appears in IEEE Transactions on Geoscience and Remote Sensing. The system can generate 3D models of entire cities using only 2D aerial photographs, automating a time-consuming manual process that previously required specially trained 3D artists and computer graphics programs. "Think about all the time and labor involved in manually creating a digital 3D model of New York City for a new Spiderman movie," said Kyle Gao, a Ph.D. student in systems design engineering. "With our system, it can be done using a few hundred aerial images—satellite images from Google Earth, for example—to train the model for a couple of hours in an automated process." The technology is built upon a method known as Gaussian Splatting, which uses millions of tiny ellipsoids, each with their own color and lighting detail, to automatically create 3D digital assets out of 2D aerial photographs. Credit: University of Waterloo "In the same way the human body is made up of tiny atoms, large-scale 3D objects are built from small 3D geometric ellipsoids," said Gao. "Or you can imag...