top of page
Mat A.

A Geolocation Walkthrough

The ability to determine the geographic location of a subject or object from images can be critical in intelligence. Known as 'geolocation', this involves using visual clues, metadata, and various open sources to pinpoint where an image or video was captured. Every detail can contribute as a valuable piece of the puzzle.


Before reading this blog, we recommend you check out the blog post titled Image Analysis and Verification.


This blog will be about how we can draw out the detail, of an image, and how can we identify "map markers". Put another way, what is evident in the image that would be observable from a satellite or street view? This blog will also look at how generative AI can support the image analysis process.


What's Your Objective?


The specific objective will guide the process you undertake, and may include:

  • identifying people

  • geo-locating an image

  • identifying objects

  • determining whether the image has been AI-generated or manipulated.


For the purposes of this blog, our objective will be to determine the precise location where the photographer was standing in the image below.


Image Analysis - Draw Out The Detail


To begin with, it's important to thoroughly analyse the entire image before drawing any conclusions. It's easy to focus on a specific feature that stands out, but doing so might lead to premature assumptions. By carefully examining all aspects of the image, we can ensure that no critical details are overlooked, which might either support or challenge our initial thoughts.

busy street scene
Geolocate This Image!

The image presents a busy street scene, with several key elements that can help us identify the location. As detail is observed, note what they are, but also what that might mean. We have listed several aspects, followed by questions to support your thinking process.


Traffic and Vehicles

  • The image shows a variety of vehicles including cars, buses, and taxis.

  • A red compact vehicle commonly found in urban environments.

  • Shadow of the vehicles.

Can we search the buses and taxis as they appear to have distinct designs and markings (which can be region-specific)? What make/models are the vehicles? Are these vehicles only sold in certain areas? Does the shadow help to pinpoint the time of the day?


Buildings, architecture and natural features

  • The architecture ranges from modern high-rise buildings to older, colonial-style structures.

  • There’s a mix of commercial and possibly residential buildings.

  • A distinctive building located on the left side of the image.

  • A mountain range visible in the background.


What does the combination of modern and older colonial-style buildings mean? What cities are near mountain ranges and does this help to narrow our search?


Pedestrian activity

  • There’s a considerable number of pedestrians, some with umbrellas, suggesting a sunny day.

  • A police officer / traffic police

  • Presence of facial masks.


What can the attire provide in terms of cultural and climatic hints? Do the masks indicate a particular time frame? Can the uniform be identified?


Traffic lights and road markings

  • Overhead traffic lights and specific road markings.

  • The presence of a pedestrian crossing, and the design of traffic lights are notable (think what will be observable on street view).

  • Two-way traffic

  • A small median strip running down the centre of the road.


Can the design of traffic signals and road markings be compared with standards used in a specific country?


Language, signs and flags

  • Some signs are visible, which can provide clues based on the language used and the names of businesses or streets.

  • Some text on signs could be discerned,

  • A flag is visible on one of the buildings, which could be a national or local flag, giving a significant clue.


If the signs are in Spanish, this could suggest a location in a Spanish-speaking country? What country/organisation does the flag belong to? If the flag is Peruvian, does that mean it is in Peru, or an embassy in another country? What does the text on the signs say, and what is being advertised? Can we search these items?


Assumptions, so far.


Based on the visual clues, we can with some confidence, assume:

  • The image has been taken in a busy urban centre in a Spanish-speaking country, possibly in Latin America.

  • The architectural styles, types of vehicles, and pedestrian activity all point toward a city with a blend of historical and modern influences.


Now, let’s analyse key elements of the image to either confirm or disprove our assumptions and further refine our search area.


Reverse Searching


Of the individual elements mentioned above, public transport as this is often linked to a specific region or city. First, we attempted to identify the white bus in the centre of the image using reverse image searching tools like Google Lens. While this did not reveal the actual location, it did suggest that the style of bus is likely found in South or Latin American countries. We then can narrow down to a city by focusing on the white taxi, which originates from Lima, Peru.


Google Lens to identify the taxi
Using Google Lens to identify the taxi from Lima

While it’s tempting to focus and go all-in on Lima, it would be prudent to look at other elements in the image to increase our confidence.


  • The flag visible on the buildings to the left. Is it the Peruvian flag?

  • What language visible in the image? Is it Spanish?

  • What style is the architecture? It appears to be in a Spanish colonial style.


But let's focus on what uniform is the police/traffic officer wearing? A quick, easy search for traffic officers in Lima revealed an exact match for the uniform.


Can we use ChatGPT to help?


Now, let’s pivot, and see if ChatGPT 4o (the paid version) gives us a similar result, or completely different. How does this compare to our findings? What are is strengths and weaknesses?


When using AI, it’s important to guide it in a way that aligns our objectives. This can be achieved by using a well-constructed prompt. To learn more about prompting AI, please read our blog post, Prompting AI for Open-Source Intelligence.

ChatGPT
Using ChatGPT 4o (17-Sep-2024) to help with drawing out the detail of the image.

As seen in the example above, the AI has produced results similar to us...thankfully!


We uploaded the image and used the following prompt:

Looking at this image, conduct an analysis to assist me in working out which country or city the photo is in. For example, a license plate number could tell you what country you are in, or a store’s phone number area code might help indicate a specific city or area. I want you to clearly articulate the items, things, objects that would help to identify the location. Focus on the following areas: Signage, such as text on billboards, shop fronts, signposts and banners. Focus on shops that appear to be independent as these often bring up only one business address. Look for phone numbers and advertisements and websites displayed on window and storefronts. Landscape, noting any identifiable features such as mountain ranges, rivers, bridges, trees or sandy terrain. This will me to have points of reference to check against satellite imagery or compare with any other geo-located images. Understand the types of roads and paths that may be specific to a given country, state or city. Structures, look for statues, fountains, water towers or buildings to help match the location with satellite imagery. Note architecture to help you narrow down your location search to specific areas. Also look at image backgrounds, e.g. room designs, to see if there is valuable information. Vehicles, look for registration details / number plates and the type, model of the vehicle. Look for items in the reflection of the windows and mirrors. Clothing, such as insignia and logos and types of clothes worn by individuals. Also, look for hand-held objects, such as phones and documents. Go through each of the above and note what is found. If nothing is found against the suggested, make a note. Once the above is complete, consider other relevant information, such as product brandings, newspapers, languages, electrical power points. Then, I want you to use your knowledge and identify anything that might be used to identify the location. Finally, draw a conclusion on where the image is, reminding the me to verify the information that you have presented!

Many of its findings align with ours, and in some cases, it has conducted a more detailed analysis. One notable example is the signage, which the AI was able to translate from Spanish to English, without added prompting. However, as with any information we find online, these results need to be verified to ensure we are not heading in the wrong direction.


You may notice that the AI incorrectly assessed the terrain as flat, with no noticeable mountains or specific natural features in the background. While the mountains are clearly visible to the human eye, the AI in this instance was unable to detect this, without specific follow up prompting. Therefore, while AI systems are a supremely valuable tool, it should be used alongside our initial analysis as an aid (one tool in your toolbox), rather than being solely relied upon. This also shows how these tools work best with constant interaction and back-and-forth refinement.

ChatGPT
Using AI Systems requires a partnership to get the best results

Image > Mapping


With the likely city identified, our next step is to examine a map of Lima. Given the large area we need to cover, we want to refine our search by using the other clues within the image. The image contains several key reference points that might be identifiable on the map, including:


  • A mountain range visible in the background.

  • A congested intersection that appears to be a main thoroughfare leading toward the mountain range.

  • A distinctive building located on the left side of the image.


By examining a topographical map (Google Maps) of Lima and utilising the terrain feature, we can view the city's mountain ranges and other natural landmarks. This approach makes it easier to pinpoint the mountain range visible in the background of the image. Considering the direction of the road, we've identified three potential areas in Lima that warrant closer focus.


Lima Map
Areas of Lima to focus on

Now that we have several areas to focus on, we can start investigating another one of our key reference points. While we could look at all the major thoroughfares approaching the right direction, this may take some time as there appear to be multiple options. As such, we will focus on the distinctive building to try and identify it or, at worst, a list of buildings it could be.

A potential key reference point in Lima
A potential key reference point in Lima

Google Lens was initially used to identify the building but was unsuccessful, likely due to the image quality. After cropping the original image, a second attempt with Google Lens yielded better results.

the Cathedral of Lima
The Cathedral of Lima

Having identified a second reference point, we can now locate this building on our map and verify if it aligns with the one in our image. The Cathedral can now be used to further narrow down our search area. We can now turn our attention to our third reference point, which is the intersection of the busy thoroughfare. We know the road is running in a direction towards the mountains, and the unique building (Lima Cathedral) is to the left of the photographer. From here, we identify major roads moving towards the mountain and in the vicinity of the cathedral.

View of the mountains and street angles
Roads in vacinity of the Cathedral heading in the direction on the mountain range

We've identified six possible roads where our intersection could be located. We can rule-out all but two, as they do not meet the specific criteria: a major thoroughfare with two-way traffic, an intersection with overhead traffic lights, and a small median strip running down the centre of the road.


We are now looking at the thoroughfare Av. Abancay. We can now attempt to pinpoint our intersection by focusing on roads that head towards the Cathedral and intersect with Av. Abancay.


Target road with possible intersections
Target road with possible intersections

With two potential intersections identified, we switch to a satellite view of the map to examine specific road features, such as the zebra crossing. Upon closer inspection, we observe that the more southerly intersection has a slightly different crossing pattern and likely offers a restricted view of the cathedral due to the density of nearby buildings. Therefore, we will now focus our attention on the intersection of Av. Abancay and Jr. Huallaga.


Now, we utilise Google Street View to confirm that we've identified the correct intersection and to see if we can match the perspective of the photographer. Using Street View, we can now identify nearly all the elements that we listed.

Street view confirmation of intersection
Google street view confirmation of intersection

With the objective of locating the image successfully achieved, the next step may involve timestamping the image to determine the approximate date and time it was taken.


The method outlined above to identify the location of the target image is just one approach to achieving this task. Using AI, or Google Lens can help to identify objects and their location but do your best to verify these data points. Whatever aspect you decide to concentrate on, it's crucial that if you form assumptions about where the image is located, you need to verify them. Be prepared for the possibility that your assumptions may be incorrect, and actively seek out information that not only supports your assumptions but also challenges them.


To support OSINT collection and analytical capability uplift and to delve deeper into some of the learnings above, we would love to talk more about NexusXplore, an all-in-one, investigation-agnostic software platform, or our various training courses. Alternatively, contact us at info@osintcombine.com for a friendly chat.

1,000 views

Recent Posts

See All
bottom of page