colstoThis video examines the amount of location information carried by inaudible background sounds, using the Doppler effect to locate a reference recording within ~100m, and only needing (in some cases) less 1-min of audio. This video raises awareness of these capabilities and closes by discussing preventative measures one could take.
Methodology: Aircraft position, altitude, and velocity data were sourced from ADS-B data and filtered in Python using ICAO type-code classification for propeller vs jet types. For each candidate location on a 200m grid, the closest point of approach (CPA) and sound propagation delay were computed from 3D slant distance, using a digital elevation model (DEM).
Doppler curve spectrogram measurements constrained each event's velocity and CPA slant distance. Candidates were then filtered by the 920-second offset between two flyovers, plus a silence constraint. The exact Doppler curve each candidate would hear was modelled, using emission-time radial velocity from interpolated ADS-B data.
Modelled doppler curves were scored against the observed spectrogram using a locked two-event scoring method. This corrected for measurement error on CPA time & fundamental frequency, while preserving the purely physics-derived temporal offset between the two events.
The pipeline was repeated again for a single 1-min clip of known time, US-wide search, and again, successfully scored the true location 1st.
Lastly, preventative measures were assessed, including rationale behind why common signal processing fixes fail (high-pass filtering, drowning out, and phone call compression/noise reduction).
This video is an educational demonstration of acoustics, signal processing, and data analysis using publicly available data. No code or tools are being released.
I discovered a sound that leaks your exact location (and its everywhere)colsto2026-11-02 | This video examines the amount of location information carried by inaudible background sounds, using the Doppler effect to locate a reference recording within ~100m, and only needing (in some cases) less 1-min of audio. This video raises awareness of these capabilities and closes by discussing preventative measures one could take.
Methodology: Aircraft position, altitude, and velocity data were sourced from ADS-B data and filtered in Python using ICAO type-code classification for propeller vs jet types. For each candidate location on a 200m grid, the closest point of approach (CPA) and sound propagation delay were computed from 3D slant distance, using a digital elevation model (DEM).
Doppler curve spectrogram measurements constrained each event's velocity and CPA slant distance. Candidates were then filtered by the 920-second offset between two flyovers, plus a silence constraint. The exact Doppler curve each candidate would hear was modelled, using emission-time radial velocity from interpolated ADS-B data.
Modelled doppler curves were scored against the observed spectrogram using a locked two-event scoring method. This corrected for measurement error on CPA time & fundamental frequency, while preserving the purely physics-derived temporal offset between the two events.
The pipeline was repeated again for a single 1-min clip of known time, US-wide search, and again, successfully scored the true location 1st.
Lastly, preventative measures were assessed, including rationale behind why common signal processing fixes fail (high-pass filtering, drowning out, and phone call compression/noise reduction).
This video is an educational demonstration of acoustics, signal processing, and data analysis using publicly available data. No code or tools are being released.
Link to full recording: youtu.be/4CxVxuMVkMcI located a fans video using thunder soundscolsto2026-09-30 | I used my acoustic signal processing pipeline and data from the Norwegian weather API to localize a storm recording sent in by a subscriber.
Lightning strike timing and position data was pulled from the Norwegian Meteorological Institute's public API. This enabled higher spatial precision at ~200m vs ~3km deviation from the public Blitzortung dataset used in the prior video.
Selective lightning strike weighting was added to the scoring algorithm to prioritize strikes outside of the main cluster.
The positional accuracy was still relatively poor at 2.2km away, given the spatial precision of the dataset; this was largely due to the heavy clustering of strikes away from the listener, limiting triangulation ability. With strikes coming from multiple directions, ~500m accuracy should be feasible.
Start Learning Cyber Security for FREE with TryHackMe: tryhackme.com/Colsto Use my code “COLSTO" to get 30% off on Annual Subscription!Can you geolocate a video using background thunder sounds?colsto2026-08-09 | This video demonstrates a forensic audio geolocation methodology using background thunder sounds from a video.
Lightning strike timing/position data was sourced by logging the Blitzortung/lightningmaps.org socket feed for ~3wks. For each candidate location, slant distance and inverse-square attenuation were used to model an acoustic curve from the lightning data, using signal processing techniques in Python. This curve was then compared against the observed audio's bandpassed (20-250hz) RMS envelope using cosine similarity, and computed across all possible time offsets using FFT-based cross-correlation.
The methodology was validated blind against a known location, then tested under progressively less-constrained scenarios: (1) Full-world search with known date, unknown time, then (2) known region, ~3-wk temporal uncertainty.
The approach could be improved by using paid/commercial lightning detection data (e.g., Xweather, Vaisala GLD360, Earth Networks ENTLN) for tighter positional accuracy per strike.
Idea was inspired by Benn Jordan's video on infrasound; would highly recommend if you haven't seen it.I geolocated an animal abuse video using OSINTcolsto2026-07-12 | This video walks through a text-based semantic search methodology for narrowing a large geolocation search space.
The case study involves identifying the exact location of a short video taken somewhere in London. First, OS Open Roads dataset was used to identify the ~80k candidate T-junctions. Next, OSM was used to filter to the ~20k junctions on 2-lane roads, and Google street view panos pulled for these. Lastly, CLIP / SigLIP models were used to score against natural-language prompts describing visible features in the footage.
A two-pass architecture was used. ViT-B/16 on all 20k candidates, then SigLIP SO400M to refine the top decile with additional directional crops (e.g., railings on both sides of a crossing)
The workflow could be improved by (1) adding image-to-image matching alongside text prompts where a reference view of the target scene is available, (2) incorporating late-interaction multi-vector scoring for features that occupy small portions of the frame, and (3) offloading the 2nd rescoring run to cloud GPU compute (e.g., Lambda Labs, RunPod, Modal) to enable faster prompt iteration and larger candidate sets.
The back half of the video covers the follow-through after the geolocation was solved; reaching out to nearby vet practices, the RCVS Code of Professional Conduct disclosure exception (Ch 14), and the current state of responses from the RSPCA and the vet practices.
This video is intended for educational purposes, focusing on methodology and analytical reasoning using publicly available data.I used OSINT to find the real One Piece ocean drop locationcolsto2026-06-13 | This video is an OSINT case study demonstrating the methodology I used to locate the One Piece ocean drop.
The analysis examines how open-source intelligence and geospatial techniques can be used to narrow down an intentionally low-signal search area with almost no visible landmarks. The workflow combines bathymetric filtering, ROV operator logistics, JAMSTEC dive history, Copernicus wind and wave data, Himawari satellite imagery / cloud-field analysis, ship traffic constraints, SAR imagery, seafloor current data, coastal sightlines, open source photography, and camera-geometry reconstruction.
This video is intended for educational purposes. Any third-party footage shown is used under fair use and is transformed for analytical and educational purposes.
---
Music: “Winter - Vivaldi,” conducted by Philip Milman Source: https://pmmusic.pro/downloads/ | License: CC BY 3.0 — creativecommons.org/licenses/by/3.0Andrew Tate challenged me to find him...colsto2026-05-27 | This video presents a structured geolocation methodology developed in response to a public challenge.
The analysis examines how open-source intelligence (OSINT) techniques can be applied to visual media through the systematic interpretation of solar geometry, shadow direction, road orientation, topographic structure, mountain-profile correspondence, vegetation distribution, roadside infrastructure, satellite imagery, map data, and geoguessr-type clues.Finding iconic rap locations pt. 8 #geolocation #osint #migoscolsto2026-05-26 | ...Everyones gonna recognize it #geolocation #earth #osint #vinecolsto2026-05-17 | Finding the location of the iconic "Why are you running?" vinerip outdoor boys #geolocation #outdoorboys #osintcolsto2026-05-17 | ...How to locate (almost) any photocolsto2026-04-17 | This video is a walkthrough of a terrain fingerprinting GIS workflow used to narrow down a large search space. this geolocation methodology relies only on terrain & structure presence, making it extremely versatile and scalable, especially for low-signal searches with limited data
The case study involves determining the location of an aerial image within a search area spanning the Atlantic rainforest (~100,000+ sq mi). the workflow combines Google’s Open Buildings dataset, a 30m resolution digital elevation model (DEM), and land cover classification data to progressively reduce the search space.
The workflow could be improved primarily by (1) using a higher resolution DEM [if available for AOI] and (2) increasing precision of the terrain template/sketch used for matching
This video is intended for educational purposes, focusing on methodology and analytical reasoning using publicly available data.No one could find where this album cover was taken...colsto2025-12-31 | I took the album cover for We Don’t Trust You by Future & Metro Boomin and tried to answer a simple question: where was this photo actually taken?
This video walks through the full OSINT and geolocation process I used, pulling from both the album cover and related teaser footage.How I found the Primitive Technology locationcolsto2025-12-19 | In this OSINT breakdown, I walk through the geolocation process step-by-step, analyzing terrain, vegetation, shadows, waterways, elevation data, satellite imagery, and environmental clues to pinpoint the exact, iconic filming site that many have wondered about for over a decade
This video is for educational, analytical, and research purposes. #OSINT #PrimitiveTechnology #Geolocationwould be so pretty at night 🥺 #rap #locationcolsto2025-11-23 | Finding the location from the iconic Bobby Shmurda rap music videoi bet no one realizes #geolocation #osint #vinecolsto2025-10-29 | ...“you can’t even tell me what country I’m in” #geoguessr #osint #geolocationcolsto2025-10-28 | ...he had no idea..colsto2025-10-22 | ...The internet couldnt find this house. How I did it with just 4 photoscolsto2025-08-16 | This video is a walkthrough of a methodology for combining public geospatial datasets, satellite imagery, solar-angle analysis, and visual pattern recognition to study how environmental clues can be used in spatial investigations.
The workflow demonstrates how broad geographic assumptions can be tested and refined using non-invasive data sources such as parcel geometry, aerial imagery, historical imagery, lighting conditions, object-color detection, and land-use characteristics.
Certain details, maps, and location-specific visuals have been intentionally generalized or omitted to preserve privacy and keep the emphasis on responsible geospatial analysis.