Dashcams have gone mainstream in the Netherlands. A small camera behind the windscreen costs a few tens of euros, records every trip and automatically saves the last few minutes after an impact. After a collision that footage is worth a lot: instead of two conflicting stories, there is video. Insurers, lawyers and the police now regularly receive a video file or a handful of screenshots as part of a claim or a police report.
With the days getting shorter and darkness falling early again from October, the number of damage reports traditionally goes up. So does the pile of footage someone has to review. And that is where a question often gets skipped: was this footage actually recorded at the location named in the claim?
Why the location of dashcam footage matters
A damage claim is about a specific situation: a junction in Amersfoort, a roundabout near Zwolle, a car park in Rotterdam-Zuid. Who is liable depends on right-of-way rules, road markings and the layout of that exact spot. If the footage comes from somewhere else, the whole reconstruction falls apart.
Most of the time nothing sinister is going on. Someone attaches a clip from an earlier trip by mistake, or mixes up two files. But some claims do it on purpose. Think of footage from a real collision elsewhere being reused for an invented incident, or a staged crash filmed on a quiet street while the claim names a busy junction where the other party is easier to hold liable.
For a claims handler the difference is not always visible. A Dutch street with terraced houses, a cycle path and a few parked cars looks like thousands of others. Without visiting the address, it is hard to say with confidence whether it matches.
Built-in GPS is not conclusive evidence
Many dashcams have a GPS module and write coordinates and speed into the video file or a separate log. That seems to solve the problem, but in practice it works less well than you would expect.
Cheaper models often have no GPS at all. On models that do, the data is lost as soon as someone takes a screenshot, trims the video or forwards it through WhatsApp. What reaches the insurer is then a bare video file without a location. And when the GPS data is there, free software can edit it. A log file is a text file, not a sealed piece of evidence.
On top of that, GPS can drift tens of metres in a city with tall buildings or in a tunnel. For a claim where the difference between two junctions in the same neighbourhood matters, that is too coarse. Just as with EXIF data in photos, metadata is a clue, not proof.
How AI geolocation checks dashcam footage
Visual geolocation ignores metadata and looks at what is actually in the frame. GeoPin compares a frame from the video with an index of millions of street-level images covering the whole of the Netherlands. It looks at facades, roof shapes, lamp posts, the width of the cycle path, the colour of the asphalt, rows of trees and how buildings sit relative to each other. The result is a predicted location with coordinates and a confidence score.
For dashcam footage this works as follows: the handler takes a few frames from the seconds before the collision, ideally moments where the surroundings are clearly visible. Those frames go through GeoPin. If the predicted location matches the address in the claim, that is a strong confirmation. If the system points to a spot several kilometres away, that is a reason to look at the claim more closely.
Video has one advantage over a single photo: you have many frames. If five consecutive frames all point to the same street, the result is far more reliable than one prediction on its own. And if frames point to different places, that can indicate the video has been edited together.
A practical example
Say a driver reports a collision at a junction in Utrecht-Oost. According to the report, the other party failed to give way, and the dashcam footage shows a van pulling out from the right. The other party denies it and says they were not even in Utrecht that day.
The claims handler pulls four frames from the video and runs them through GeoPin. Three frames point with high confidence to a junction in Nieuwegein, more than five kilometres away. The terraced houses have a different type of gutter than those on the stated street, and the cycle path is red asphalt instead of paving bricks. That is not a final verdict, but it is a concrete reason to ask the reporting driver for clarification and perhaps request the other party’s footage.
The reverse happens just as often: the other party doubts the footage, GeoPin confirms the location, and the argument about where it happened is settled in a few minutes. That saves sending an expert to the scene.
Limitations: night, rain and motorways
Not every frame is equally useful. Many autumn collisions happen in the dark or in the rain, and then the surroundings are hard to make out on dashcam footage. Headlights cause glare, wipers sweep across the image and distant details disappear. We wrote earlier about the challenges of night photography for geolocation. The rule of thumb: pick frames where street lighting, facades or traffic signs are visible.
Motorways are a different story. A stretch of the A2 with noise barriers and crash barriers looks a lot like a stretch of the A12. Street-level geolocation works less well there, and hectometre markers or exit signs are better anchors. Inside built-up areas, where most minor collisions happen, the surroundings are full of recognisable detail.
Finally, a geolocation result is never the whole story. It is a signal that helps the handler decide where to focus, not an automatic rejection. The final judgement stays with a person.
Fitting it into claims handling
For insurers and loss adjusters handling large volumes, uploading frames by hand does not scale. With the GeoPin API the check can run automatically: as soon as footage is attached to a claim, a few frames go to GeoPin and the result appears as an extra field in the file. Claims where the predicted and stated locations are far apart get flagged. Everything else carries on as normal.
GeoPin looks at the surroundings, not at people: the model is not built to recognise faces or number plates. That keeps processing limited to what is needed to check the location, which fits the GDPR requirements insurers have to meet.
Try it yourself
Have dashcam footage or a screenshot of a traffic situation and want to know whether the location is right? Upload a frame on GeoPin and see within seconds where the model thinks it was taken. If you work at an insurer or claims firm and want to build this into your own process, read how GeoPin works or get in touch about API access.