boxmot.engine.calibration.kalman_model_3d

Unscaled EagerMOT covariance bases for supervised 3D calibration.

CalibrationModel3D

CalibrationModel3D(options: Mapping[str, Any])

Expose the runtime box7 state and its original, unscaled P0, R, F, Q.

Geometry is bottom-center x, y, z, yaw, length, width, height in the runtime world frame. Yaw follows the filter's pi-equivalent convention. Prediction advances one delivered frame, including for angular velocity.

dim_x instance-attribute

dim_x = len(state)

dim_z instance-attribute

dim_z = 7

measurement_indices instance-attribute

measurement_indices = tuple(range(dim_z))

velocity_indices instance-attribute

velocity_indices = tuple(range(dim_z, dim_x))

velocity_measurement_indices instance-attribute

velocity_measurement_indices = (
    (0, 1, 2, 3) if dim_x == 11 else (0, 1, 2)
)

initial_state

initial_state(
    measurement: ndarray,
) -> tuple[ndarray, ndarray]

Return observed geometry, zero velocity, and the runtime birth prior.

measurement_covariance

measurement_covariance(
    measurement: ndarray, score: float = 1.0
) -> ndarray

Return the source constant R, independent of detector confidence.

process_covariance_bases

process_covariance_bases(
    mean: ndarray, dt: float | None = None
) -> tuple[ndarray, ndarray]

Separate the full measured-state and velocity blocks of runtime Q.

to_measurement

to_measurement(
    box: ndarray,
    score: float = 1.0,
    reference: ndarray | None = None,
) -> ndarray

Validate box7 geometry and align yaw exactly as a runtime update.

transition

transition(dt: float | None = None) -> ndarray

Return the fixed-frame runtime transition matrix.