Kalman Filter MCQ
Recursive Bayesian estimation for linear systems with Gaussian noise—predict with motion model, correct with measurements (e.g., box center).
Predict
Model
Update
Measure
Noise
Q, R
State x
pos, vel
Kalman filter
For linear dynamics and Gaussian noise, the Kalman filter is the optimal recursive estimator. In CV, constant-velocity models smooth bounding boxes or keypoints between detections.
Two steps
Predict: propagate mean and covariance with F, Q. Update: fuse measurement with H, R via Kalman gain.
Details
State vector
Often position+velocity per axis; higher order adds acceleration (constant acceleration model).
Q vs R
Process noise Q allows model mismatch; measurement noise R trusts observations.
Nonlinear
EKF/UKF/particle filters extend when motion or measurement models are nonlinear.
In tracking
Prediction gates association; update pulls estimate to matched detection.
Recursion
x̂_{k|k-1} → z_k → x̂_{k|k}