Which AI-supported geolocation verification tool works best?
,Geolocating digital media is a tedious job. Do we really still need to put so much manual effort with all these AI-supported tools that are available nowadays? We have tested Overpass Turbo, GeoGuessr GPT, GeoSpy, Earthkit and SPOT to help you with your geolocation challenges. Here's what we have learned.
Next to the robust but quite techy Overpass Turbo, there are a couple of publicly available geolocation solutions that harness the power of LLMs. In our opinion, GeoGuessr GPT, GeoSpy, and EarthKit are among the most promising tools currently available. To demonstrate and discuss the advantages of and differences between these tools and our self-developed SPOT-application, let’s look at an example and compare.
The example is an image from the (excellent!) Overpass Turbo (OT) tutorial. We know that this image is supposed to have been taken in the town of Legnago, Italy. Let's check that!
The most prominent visual clues we can see in this image are a parking lot, a residential building, a power tower and a communication tower, all of which are entities likely to be found on OpenStreetMap (OSM). (Look closer and you‘ll even find more, some of which are unfortunately not tagged in OSM.)
Overpass Turbo
Let’s start with a search in Overpass Turbo, developed (in 2012!) by Martin Raifer. An experienced OT user writes a query like the one in the image below. This query will lead you to a perfect single result on the map:
In the OT interface all entities that have been searched for are clearly indicated on the map. To check if the marked location matches the search, you can download or copy the coordinates into Google Maps and corroborate the results with satellite or Street View images.
Our thoughts: It is a powerful geolocation tool, but you need to understand the OSM tagging system and OT Query language, which may not be for everyone.
GeoGuessr GPT
GeoGuessr GPT, created by Bill Meixner, is a Generative Pre-trained Transformer (GPT). A GPT is a prompt setup for ChatGPT. This one is available for free and it instructs the LLM to behave like an OSINT investigator specialized in geolocation verification.
Using GeoGuessr GPT, you can upload an image and provide additional information. An example prompt could be: "Where in Legnago, Italy can I find this parking space within 200 meters from a communication tower and a power tower?".
Pasting the coordinates into Google Maps shows that the location specified by GeoGuessr is about 1km away from the actual location in this image. A couple more runs with the same prompt and image provided varying results with up to 1,4 km of distance between the actual and the suggested location.
Our thoughts: GeoGuessr GPT provides you with visual analyses and points out the geographical elements that are taken into consideration for its results. The results and suggested coordinates are mostly not accurate enough to use. Nevertheless, it’s good to use the hints if you are stuck in your geolocation challenge.
GeoSpy
Another LLM-based tool that is making the rounds in OSINT circles is GeoSpy, developed by Graylark Technologies LLC, a US-based company. GeoSpy runs visual analysis on an uploaded image and provides you with a possible location within seconds. There is no option to add any contextual or known information (e.g. a city name).
GeoSpy is amazing in recognizing (mostly rough estimates of) locations. In our example, it correctly identifies North-East Italy, and provides detailed information about detected entities, architecture, vegetation, and overall landscape. However, the results are, in most cases, indicative. Much work is still needed to find the precise location.
The disclaimer on the website might be a good indicator of where the tool currently is and soon might be: “This version of GeoSpy is a demo in beta and may produce inaccurate results. It should not be used for any serious purposes.”
Our thoughts: Geospy is very helpful when you have no idea where to start.
EarthKit
Earthkit, developed by Jett Chen allows for different search modi. One of them is a geo estimation service that works just like GeoSpy with similar indicative results. Again, Northern Italy is predicted correctly but from there on it takes much manual work to come up with a precise location.
The option to search by ‘Overpass Query’ is more precise. It provides a way to help you write an OT Query Language prompt, using @... to tag locations (such as @Legnago) and #... to indicate ‘features’ (such as #electricity which leads you to power=tower).
Our thoughts: Earthkit’s 'Overpass Query' is definitely an improvement over writing prompts in OT Query Language yourself. However, a good understanding of the OSM entity tagging and relation structure is still required to come to relevant results. The interface of Earthkit does not support user needs like zooming, map layers, saving & sharing results, etc.
SPOT
Lastly, we take a look at SPOT, developed by us, DW Innovation. SPOT is an LLM-based geolocation service that works with natural language prompts. It requires no coding skills. Just type what you are searching for, in this case: “a parking space within 200 meters from a communication tower and a power tower in Legnago, Italy”.
Spot semantically understands and translates your prompt to an OSM search and displays the search results on a map. The interface allows you to check the location on Google Street View or other map services like Bing and Yandex. The application is still in Beta, so you sometimes need to change or rephrase a prompt for SPOT to understand your search.
Our thoughts: SPOT performs well and in multiple languages. The results are relevant and easily checked because of the integration of Street View. Our aim is to provide you with a tool that significantly simplifies location verification. Try SPOT at findthatspot.io.
Which geolocation tool works best?
All tested AI supported tools provide a location and are helpful in doing geolocation work, even if they only help you to find a starting point. One day, GeoGuessr GPT, GeoSpy, and Earthkit, might get to a level where they can tell you the exact coordinates of a location shown in an image or video. Until then, inaccuracies and uneasy handling leave a lot of additional work to find the exact location. With SPOT, users can efficiently work their way to an exact location. To sum up, AI supported geolocation is becoming more efficient and powerful every day!
