Geodesy has traditionally relied on instruments such as GNSS receivers, total stations, radar satellites, and laser scanners to measure Earth's shape and how its surface changes. But another technology is rapidly becoming part of the geodetic toolbox: computer vision.
Computer vision allows computers to extract measurements and patterns from photographs, satellite imagery, and video. Combined with artificial intelligence, it can transform huge collections of images into information about surface movement, elevation, glaciers, earthquakes, and infrastructure deformation.Turning Photographs Into 3D Terrain
One of the most useful applications is Structure from Motion (SfM) photogrammetry. Instead of directly measuring millions of points, researchers take overlapping photographs of an area from different positions.
Computer-vision algorithms identify matching features between photographs and use their changing positions to calculate depth. Thousands of images can therefore be converted into dense 3D point clouds, digital elevation models (DEMs), and orthomosaics.
Drones make this especially powerful. Researchers can repeatedly fly the same route over a landslide, glacier, coastline, or fault and construct a new 3D model during every survey.
Comparing those models reveals changes that would be difficult to recognize from photographs alone.
Watching Glaciers Move
Glaciology is particularly well suited to computer vision because glacier surfaces contain recognizable features such as crevasses, debris, and cracks.
Feature-tracking algorithms can identify the same surface patterns in satellite images taken days or weeks apart. Their displacement between images reveals how far the ice moved.
This creates glacier velocity maps covering enormous regions without requiring instruments directly on the ice. Similar techniques can track changes in glacier termini and automatically identify expanding or shrinking glacial lakes.
Machine-learning systems are also becoming increasingly capable of separating snow, ice, rock, water, and shadows in satellite imagery. This could make monitoring thousands of glaciers considerably faster.
Measuring Earthquakes from Images
Computer vision can also help geodesists study earthquakes.
When an earthquake shifts the ground horizontally, features such as roads, river channels, fields, and ridges can move between pre- and post-earthquake satellite images. Image correlation measures these offsets across thousands of locations.
The result is essentially a displacement map created from photographs.
These measurements complement GNSS and InSAR. GNSS provides extremely precise movement at individual stations, while InSAR measures deformation using radar phase differences. Optical computer vision can provide another perspective, particularly where very large ground movements make radar measurements difficult.
AI Could Make Geodetic Monitoring Continuous
Perhaps the most interesting emerging application is combining computer vision with the enormous volume of imagery now produced by Earth-observing satellites.
Traditionally, researchers might manually select images and analyze a specific glacier or landslide. An AI-based system could instead continuously examine new imagery and automatically flag unusual changes.
A model might detect a rapidly accelerating glacier, newly forming glacial lake, expanding landslide, coastal erosion, or surface rupture and direct researchers toward the event.
This creates the possibility of moving from periodic surveying toward near-continuous geodetic monitoring.
Smartphones May Become Geodetic Instruments
Computer vision could even expand geodesy beyond specialized scientific equipment. Modern smartphones contain high-resolution cameras, GNSS receivers, accelerometers, gyroscopes, and, in some models, LiDAR sensors.
Researchers are experimenting with using these sensors to construct local 3D models and measure structures. Although smartphones cannot replace survey-grade GNSS or terrestrial laser scanners for the highest-precision measurements, they could make rapid mapping dramatically more accessible.
The future of geodesy may therefore involve an increasingly powerful combination:
GNSS provides position. Radar measures deformation. LiDAR measures geometry. Computer vision interprets what is changing.
Together, these technologies could allow scientists to observe Earth's surface at a scale and frequency that traditional surveying alone could never achieve.