Recentering OpenSky Traffic: Arbitrary-Center, Multi-Projection Visualization on a Single GPU
Poster presentation at the 14th OpenSky Symposium, Madrid, 29–30 October 2026. The accompanying short paper has been submitted to the Journal of Open Aviation Science.
This page is the web companion to the poster. 日本語版 →
Aviation asks: "From this airport, in which direction, and how far?" The Mercator projection preserves local angles, but preserves neither distance nor bearing. The azimuthal equidistant projection preserves both — measured from the single point placed at its center. That is why putting the airport at the center matters. AAE can place any point on Earth at the center of any of its 20 projections, driven by the exact same underlying data.

Range rings every 2,000 km form true circles. Bearing and distance from the airport can be read directly. The outer rim is the antipode (20,000 km).

The same four routes. Rhumb lines (dashed) appear straight, but high latitudes are stretched and distance ratios break down.
Routes: MAD–DXB, DXB–HND, MAD–HND, and MAD–JFK. Solid = great circle (shortest path), dashed = rhumb line (constant bearing). The conference poster overlays these routes with 44 Spanish-registered aircraft from a single OpenSky snapshot (23:27 UTC, 23 September 2026). Coastlines: GSHHG / Airports: Natural Earth
Airports are specified by three-letter IATA codes (MAD, DXB, HND, JFK), with coordinates resolved from the Natural Earth airports dataset. Lines are computed geometrically on the sphere and projected through the same forward pipeline as point data, ensuring consistent clipping across all projections.
| Line | Meaning | Straight (or true circle) in |
|---|---|---|
| Great circle | Shortest path between two airports on the sphere | Gnomonic (all); Azimuthal equidistant (only routes through the center) |
| Rhumb line | Path of constant compass bearing | Mercator (standard orientation) |
| Range ring | Locus of equal distance from one airport | Azimuthal equidistant (forms a true circle only for the airport at the center) |
In an azimuthal projection, any great circle through the center is straight by construction, so routes radiating out of Madrid appear straight. However, a route that does not pass through Madrid remains curved: on a Madrid-centered azimuthal equidistant map, the Dubai–Tokyo route (9,240 km) bows outward from its straight-line chord by up to 1,050 km (11% of its total length). Choosing a projection means choosing which geometric question to answer.
| Question | Projection | Preserves | Cost |
|---|---|---|---|
| Distance and bearing from here | Azimuthal equidistant | Both, measured from the center | Shapes distort increasingly toward the rim |
| Is the shortest path straight? | Gnomonic | All great circles are straight | Limited to within 60° of the center |
| Which region is larger? | Equal Earth | Relative area | Local angles and shapes are not preserved |
| Constant-bearing course | Mercator | Local angles (rhumb lines are straight) | Preserves neither distance nor area |

Area is preserved. Routes curve and range rings deform into ellipses, but comparisons of regional and airspace size become accurate.
Observations from N sources are normalized into a single record type, rotated on the sphere by R, and sent to M projections.
A 40-second clip of the azimuthal equidistant projection as its center moves continuously from Madrid to Tokyo. Range rings remain fixed circles on the globe but deform dynamically across the map plane. The second clip overlays the same motion with positions of aircraft, ships, and ground stations at a single timestamp. Silent; captions in English and Japanese (enable CC).
Re-projecting raw vector polygons from scratch for every new center proved too fragile for interactive use: seam crossings and pole enclosures shift continuously with the center, and six iterations of a forward polygon rasterizer failed to eliminate topological edge cases. Instead, AAE bakes base geometry once into a static equirectangular intermediate sheet whose seam is fixed at ±180°. For any chosen center and projection, an inverse look-up table (LUT) maps each output screen pixel back to its source coordinate (inverse project, rotate by Rᵀ, and sample the sheet). Moving the center simply updates the rotation matrix R and rebuilds the LUT. Dynamic point observations and route vectors travel the opposite way, forward-projected using the same R.
| Item | Value | Conditions |
|---|---|---|
| Center change to frame ready | 16.4 ms (median) | including 6.0 ms for LUT generation (at 1,920 × 1,080) |
| 100,000 points drawn | 3.1 ms | at 1,920 × 1,080, median of 20 frames |
| Projections registered | 20 | no per-projection branches; dispatched via declared attributes |
| GPU | Single NVIDIA RTX 4000 SFF Ada | CuPy + custom CUDA kernels for scanline fill, sampling, and compositing; no graphics API |
| Item | Details |
|---|---|
| Venue | 14th OpenSky Symposium, Madrid (CRIDA), 29–30 October 2026 |
| Format | Poster (Poster Session, 29 October, 16:20–17:20) |
| Short paper | Submitted to the Journal of Open Aviation Science (4 pages); a link will be added upon publication |
| Author | Independent Researcher, Tokyo |
| Contact | mapshift.dev [at] gmail.com Please put "AAE" in the subject line. |
| Keywords | ADS-B visualization / OpenSky Network API / map projection / GPU-accelerated visualization / azimuthal equidistant |
| Data | Source | Terms |
|---|---|---|
| Aircraft state vectors (ADS-B) | The OpenSky Network | Non-commercial research terms; underlying raw records are not redistributed |
| Airport coordinates | Natural Earth 10m airports v5.0.0 | Public domain |
| Coastlines / land | Natural Earth 110m; GSHHG 2.3.7 (Wessel & Smith) | Public domain / LGPL |