Geolocating a random island using geometry and CUDA programming
Article URL: https://yassa9.github.io/osint/gralhix-004/ Comments URL: https://news.ycombinator.com/item?id=49360545 Points: 317 # Comments: 55
On August 16, 2026, reporter Sofia Santos described a challenge presented as gralhix 004, which involved geolocating an island using geometry and CUDA programming. The goal was to identify a specific resort located on an island by analyzing a photo. The process began by examining the image metadata, but it was found to be useless as there was no EXIF data, GPS information, or camera details.
The solution involved estimating the relative distances between three landmasses in the photo and calculating the angles of the triangle formed by them. A small click GUI was created to record pixel coordinates for each point in order, with a ±20% tolerance band around both values to account for precision. The dataset used for this task was OpenStreetMap's split land polygon set, which contained detailed coastline vectors in WGS84.
The filtering process involved discarding landmasses outside the tropics, as the resort islet was identified as tropical. Out of 141,131 land polygons that survived the filter, each surviving point was analyzed for nearby points within a 20km radius. If a point had at least three neighbors within this range, it was considered a cluster.
From each cluster, combinations of three points were generated to form candidate triangles. Due to the potentially large number of combinations, each cluster was capped at 60 points, selected by size rather than randomly. Each candidate triangle was then processed in parallel using CUDA threads. The threads sorted the points by land area to identify the smallest point (in this case, the resort islet), P0.
Using the winding direction of the other two points, threads assigned P1 and P2. The angle at P0 and the distance ratio were calculated independently for each thread. The threads then checked if the angle, ratio, P0's size, separation between P0 and P1, and the lengths of the sides all fell within the predefined tolerance windows.
After passing the initial mask, a second test was performed to ensure that the space next to the candidate triangle was open water. A rectangle was built along the P0→P1 edge, and the land dataset was checked for any obstructions within that rectangle. Additionally, the Polsby Popper Score was calculated to determine if the shape of the resort islet resembled a coral cay, with a score below 0.5 indicating a more jagged or elongated outline.
Finally, a minimum rotated rectangle was used to analyze the aspect ratio of P0's polygon, ensuring it matched the expected characteristics of a coral cay.
Written by urgent.news from Hacker News Best's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
This story
This is one outlet's version. Read the fullest account.
- Geolocating Random Islet Image Using Geometry & CUDA GPU Programming yassa9.github.io