Skip to content

kalbee

kalbee logo


kalbee is a clean, modular Python toolkit for filtering and tracking. It provides a unified interface for 18 filter types, smoothers, multi-object trackers, diagnostic metrics, and a built-in experiment runner to compare filter performance.

Highlights

Category What you get
18 Filters KF, EKF, UKF, SigmaPointUKF, CKF, PF, RBPF, EnKF, Information, ABG, Adaptive, Square-Root/Cholesky, Vectorized, Fading-Memory, H∞, IMM, VB, InEKF
Sigma Points Simplex, MerweScaled, Julier — pluggable via Strategy pattern
Motion Models Ready-made constant-velocity, constant-acceleration, and coordinated-turn \((F, Q)\) builders
Tracking SORT-style multi-object tracker with Hungarian association and gating
Learning Offline EM to fit \(Q\)/\(R\) from data, plus NIS-based auto-tuning
Smoother Rauch-Tung-Striebel (RTS) backward smoother
Diagnostics RMSE, NEES, NIS, Log-Likelihood, FilterDiagnostics, consistency tests
Outlier Rejection Chi-squared gating, Mahalanobis gating, adaptive outlier detection
Experiments One-liner to compare filters on synthetic signals
Stability Joseph form covariance updates, Cholesky factor stabilization, symmetry checks
Utilities Batch processing, state serialization, control inputs, missing data handling

See It In Action

Kalman filter smoothing a noisy signal
A Kalman filter turning noisy measurements into a clean position and velocity estimate, with the ±1σ uncertainty band shrinking as measurements arrive.

Animated demos for filtering, maneuvering targets (IMM) and real-pedestrian multi-object tracking live in the Examples & Gallery.

Quick Start

pip install kalbee

from kalbee import run_experiment

# Compare filters on a sine wave
report = run_experiment(
    signal="sine",
    filters=["kf", "ekf", "ukf", "pf"],
    noise_std=0.5,
)
print(report.summary())