Target Tracking — Kalman & Extended Kalman Filter

KF: optimal for linear Gaussian
x̂ = Fx + Bu, P = FPF' + Q
K = PH'(HPH'+R)⁻¹

EKF: linearizes h(x) via Jacobian
h(x) = [√(x²+y²), atan2(y,x)]
H = ∂h/∂x (Jacobian)

Process noise q: too small → lag on turns; too large → noisy estimates
Wrong model (CV on turns) → large RMSE