A blurry surveillance video showing little more than a line of trees may offer few visual clues about where it was recorded. But faint aircraft sounds in its audio helped videographer Colsto narrow down the camera’s location, according to a demonstration he published on November 2, 2026.
Colsto began with the recording date and a search area covering the contiguous United States, but did not know the time. He examined the audio using a spectrogram, which displays sound frequencies over time. The resulting curves revealed changes associated with the Doppler effect: as an aircraft moves toward and then away from a stationary microphone, its sound shifts in pitch.
Those changes can help estimate an aircraft’s speed and its closest distance to the microphone. Colsto identified two propeller-aircraft passages about 15 minutes and 20 seconds apart, then compared them with ADS-B flight-tracking data for flights crossing the United States on the recording date. The data includes aircraft position, altitude and speed.
He calculated where each flight might have been heard, accounting for the time sound takes to travel and for aircraft altitude and terrain elevation. The lack of other detectable aircraft between the two passages helped eliminate locations where another flight should have been audible. This reduced the number of possible sites from about 73,000 to roughly 15,000.
Colsto then compared the recording’s sound patterns with simulated audio based on known flight paths. After refining his search grid, a particular combination of location and flights stood out. The analysis narrowed the site to an area about 100 meters across. He used Google Street View to compare the trees in the area with those in the video and identify the camera’s location.
In a second test, Colsto used just one minute of audio containing another aircraft. This time he also knew the recording’s time: 20:37 UTC. Restricting the search to flights at that moment again led him to the same area, showing how a known time can make a single aircraft passage useful in this analysis.
The demonstration required several months of work, more than 5,000 lines of code and handling 50 GB of data. It depends on usable aircraft noise, available flight data and information that limits the search. Colsto also tested audio filtering and transmission through a phone call, but those experiments do not establish that every filter or call would preserve a usable signal.
The method does not mean that any video can be geolocated quickly. Colsto described it as an unlikely way for someone to track an individual, and the demonstration does not measure success across a large collection of everyday videos. It does show how audio may provide clues when images alone reveal little, particularly in open-source investigations.
Sources: Numerama.