Drone Photogrammetry and Computer Vision: The New Face of Land Mapping

Not long ago, producing a detailed map of a farm, construction site or village required survey teams to walk the land for days with heavy equipment. Today, a small drone can fly over the same area in less than an hour and capture hundreds of overlapping photos. Computer vision software then stitches these photos into a single, accurate aerial image.

This combination of drones and image processing, known as aerial photogrammetry, is changing how governments, engineers and farmers understand land. It is faster, cheaper and often safer than traditional surveying.

What Is Aerial Photogrammetry?

Photogrammetry is the science of taking measurements from photographs. In aerial photogrammetry, a drone or aircraft captures images from above. Because each photo overlaps with the next, software can find matching points and combine them into a large, seamless map called an orthomosaic.

Common uses include:

•         Land registration and boundary mapping

•         Monitoring construction progress

•         Measuring stockpiles in mines and quarries

•         Assessing crop health in agriculture

•         Documenting damage after floods or earthquakes

The Role of Computer Vision

The heart of photogrammetry is image stitching. To stitch photos correctly, software must detect distinctive points, such as corners of buildings or road edges, and match them across images. This step is called feature detection.

Prince (2023) explains that modern computer vision builds on these classic ideas, even as deep learning has become dominant. Fast and reliable feature detectors are still essential because drones capture thousands of images and every second of processing time matters. Géron (2022) also notes that combining traditional image processing with machine learning often gives the most practical results in real projects.

Popular feature detection approaches include:

•         FAST (Features from Accelerated Segment Test): very quick corner detection, suitable for large image sets

•         SIFT and SURF: robust to scale and rotation, but heavier to compute

•         ORB: a fast, free alternative that combines FAST with a descriptor

Research from Telkom University

A study by Sardi, Yulianto and Sthevanie (2024) analyzed how the FAST corner detector performs when stitching images for aerial photogrammetry. Their findings give useful guidance for practitioners:

•         A FAST setting of n = 9 produced the largest number of corners and was selected as the best configuration

•         The detector reached 100% precision on all nine test shapes, matching the corners that humans identified

•         Stitched images had a cross-correlation of about 0.97 with the original images in the best cases

•         Accuracy dropped for rotated images (about 0.88) and scaled images (about 0.75)

•         In a test with 20 respondents, most stitched images were rated "Excellent"

These results show that FAST is effective and fast, but drone operators should keep flight paths consistent. Large changes in rotation or altitude between photos can reduce stitching quality.

Practical Tips for Better Drone Maps

Based on current practice and research, teams can improve results by following a few simple rules:

•         Plan flights with at least 60–80% overlap between photos

•         Fly at a steady altitude to avoid scale differences

•         Keep the camera angle consistent to limit rotation

•         Capture images in good, even lighting

•         Use ground control points when high positional accuracy is required

•         Check stitched results for gaps or distortions before field teams leave

Benefits for Different Sectors

Government and land administration can update maps quickly, which supports land certification and urban planning. Construction firms can compare weekly aerial images to track progress and detect problems early. Agriculture benefits from field-level images that reveal irrigation issues or uneven crop growth. Disaster response teams can map affected areas within hours, helping them plan rescue routes and estimate damage.

From Images to Decisions

A stitched map is only valuable when people use it to make decisions. Many organizations now combine orthomosaics with geographic information systems (GIS), allowing planners to measure areas, draw boundaries and compare changes over time. Some teams also add machine learning models that automatically detect buildings, roads or damaged crops in the final image, turning raw pictures into actionable information within the same day.

Challenges

Drone photogrammetry is powerful, but it also has limits:

•         Regulations: many countries require permits for drone flights, especially near airports or cities

•         Weather: wind and rain can disrupt flights and blur images

•         Processing power: large projects require strong computers or cloud services

•         Skills: operators need training in both flying and data processing

•         Privacy: aerial images may capture people and private property

Conclusion

Drone photogrammetry, powered by computer vision, is turning aerial mapping into a fast and affordable tool for many industries. Research from Telkom University confirms that fast corner detection methods such as FAST can produce accurate stitched images, especially when photos are captured with consistent angle and scale. As drones become cheaper and software smarter, more organizations will be able to map their land with precision that was once available only to large survey companies.

FAQ

What is image stitching in photogrammetry?

It is the process of combining many overlapping drone photos into one large, seamless aerial image.

Why is the FAST detector popular?

It finds corner points very quickly, which helps process large numbers of drone images efficiently.

Do I need a license to fly a mapping drone?

It depends on local rules. Many countries require registration or permits, so always check national aviation regulations.

References

Géron, A. (2022). Hands-on machine learning with Scikit-Learn, Keras, and TensorFlow (3rd ed.). O'Reilly Media.

Prince, S. J. D. (2023). Understanding deep learning. MIT Press.

Sardi, I. L., Yulianto, F. A., & Sthevanie, F. (2024). Analysis and implementation of a fast corner detector on image stitching in the formation of aerial photogrammetry images. International Journal on Information and Communication Technology (IJoICT), 10(2), 259–269. https://doi.org/10.21108/ijoict.v10i2.1031