Solutions — 07
AI Change Detection
A single survey flight shows a condition. Only the comparison of two flights shows what happened.
The same route, the same altitude, the same camera settings — repeated at fixed intervals. The DroniumApp overlays the imagery and reports the differences: vegetation growing towards the line. A crack that was shorter last month. A stockpile that has shifted. An area that was not built on before.
The work is not in the flying, it is in the comparing. Holding two surveys of a site against each other by hand takes hours and reliably finds only the obvious. Here the analysis is done by a model trained on your own imagery, which learns with every pass what counts as normal on your site.
Why a single survey is not enough
Have a site surveyed once and you get a picture of its condition today. That is useful for a baseline and worthless for the question actually asked in operations: has anything changed, and if so, how fast.
That question drives maintenance. A crack in a concrete abutment is not a finding in itself — concrete cracks. A crack that has grown four centimetres in eight weeks is one. Vegetation under an overhead line is normal until it is not. The decision hangs on the development, not on the condition.
And that development is exactly what common practice loses. Imagery is filed away, the next appointment is a year later, and the comparison happens from memory at best.
How the comparison is produced
The basis is a repeatable flight. Missions are planned once in the DroniumApp and then flown unchanged: same route, same altitude, same camera angles, where possible at a similar time of day. Without that repeatability you are comparing two perspectives rather than two points in time.
The imagery is overlaid with georeferencing. What stands out is not reported as an image difference — shadows, wet asphalt and a different sun angle produce hundreds of those. What gets reported is what the model classifies as a change to the object, not to the picture.
The model learns on your site. What is worth a report at a substation is everyday life on a landfill. That distinction cannot be set in general terms, only trained on your own imagery — and every confirmed or dismissed report improves the next one.
What you end up with
Not a folder of images but a list: what changed, exactly where, by how much, since when. Every entry with the before and after imagery side by side, so the assessment stays verifiable rather than having to be believed.
What of that is handed to a ticketing or maintenance system depends on your environment — the DroniumApp exposes the findings through an interface.
Where it pays off
Wherever an area has to be walked regularly and walking it is expensive, dangerous, or simply too rare. Overhead line corridors, railway sections, dyke stretches, stockpiles, construction sites in progress, extensive estates under usage conditions.
The calculation rarely turns on the price of a flight. It turns on the inspection walk it replaces, on the damage caught earlier, and on the evidence you hold in a dispute.
→
Benefits at a glance
- Changes are reported, not hunted for — the analysis runs automatically
- Comparable imagery through repeatable missions instead of chance perspectives
- The model learns on your site what is normal and what deserves a report
- Complete documentation of the progression, usable as evidence