Geometry and CUDA GPU Used to Geolocate Mystery Island
Original: Geolocating a random island using geometry and CUDA programming
Why This Matters
Demonstrates practical CUDA acceleration for geospatial OSINT searches at global scale.
Developer yassa9 solved OSINT challenge gralhix #004 — geolocating an unidentified island resort photo — using triangle geometry, OpenStreetMap polygon data (882 MB), and CUDA GPU programming instead of Google Lens reverse image search.
Developer yassa9 published a detailed writeup describing how they solved OSINT challenge gralhix #004, created by Sofia Santos, without using AI image recognition tools. The task required identifying a resort name, island coordinates, and camera direction from a single drone photo. With no GPS or EXIF metadata available in the image file, the developer extracted a geometric fingerprint from three visible landmasses in the photo, measuring relative distances and triangle angles via a custom pixel-click GUI tool (01_triangle_gui.py). The search used OpenStreetMap's global split land polygon dataset (882 MB, WGS84 format) as the candidate database. A series of heuristic filters progressively reduced candidates: a tropical latitude band (−30° to +30°) retained 141,131 polygons; a local density filter (max 10 neighbors within 5 km) cut the list to 51,576; a clustering filter requiring at least 3 landmasses within 20 km reduced it further to 23,500 clusters. Candidate triplets were then generated using combinatorics (C(n,3)), with cluster sizes capped at 60 points to control compute load. CUDA GPU programming was used to accelerate the geometry matching across this large search space. All code and a final report are publicly available on GitHub.