Research · Astrophysics · Meteor tracking

Orbital determination

Extended Kalman filter state feeds a GPU Monte Carlo ensemble: 1,000,000 particles propagate under Kepler two-body dynamics for 30 seconds. Interactive 3D plots show the true track, EKF estimate, camera geometry, and the resulting uncertainty cloud.On mobile, tap Open plot for full-screen 3D interaction.

← All research

Meteor tracking · EKF + GPU Monte Carlo

Particles
1,000,000
Plot sample
10,000
Propagation
300 × Δt = 30s
Timestep Δt
0.1 s
μ (Earth)
3.9860e+14 m³/s²
Velocity σ
100 m/s
Prior covariance σ (x, y, z)
[9256 m, 11380 m, 4556 m]

EKF posterior mean [4.321 Mm, 938.3 km, 4.504 Mm] seeds a GPU particle cloud; two-body Kepler propagation advances each state over 30 s. True final position [4.308 Mm, 945.9 km, 4.497 Mm] is shown for comparison against the spread after stochastic velocity perturbation.

Extended Kalman filter · 3D track

True orbit, EKF estimate, cameras & observations

Pinch or drag inside the plot to rotate · use full screen for easier 3D control.

Open plot ↗
Loading 3D plot…

GPU Monte Carlo · uncertainty cloud

1,000,000 particles after 30s propagation

Pinch or drag inside the plot to rotate · use full screen for easier 3D control.

Open plot ↗
Loading 3D plot…

Method

Positions are sampled from the EKF Gaussian posterior (Cholesky factorization on GPU). Each particle receives a small random velocity draw, then is integrated with explicit two-body dynamics a = −μ r / |r|³. The resulting 3D scatter shows how epistemic position uncertainty and velocity ambiguity compound over a short horizon — the core question in meteor orbital determination when observations are sparse and nonlinear.

qdit-slack