Measurements¶

A measurement defines a residual that is added to the optimization problem. They share the functionality of the following (unavailable) base class.

class kontiki.measurements.Measurement¶

error(trajectory : Trajectory) → ndarray¶: Compute the error (a ND-vector), before application of any loss function.

measure(trajectory : Trajectory) → ndarray¶: The measured quantity, e.g. IMU accelerometer, or camera image projection.

Camera measurements¶

The camera classes share the following base class:

class kontiki.measurements.CameraMeasurement(camera, observation, huber_c=1)¶

Measurement of an sfm.Observation instance, using a sensors.Camera. huber_c specifies the design parameter of the used Huber loss function.

observation¶: The sfm.Observation instance that is measured.

camera¶: The sensors.Camera used.

Camera measurement classes¶

class kontiki.measurements.StaticRsCameraMeasurement¶

Static rolling shutter projection

Projects a landmark into a rolling shutter camera by simply setting the projection time from the observed image row. This method does in general not fulfill the rolling shutter projection time constraint, but is fast.

class kontiki.measurements.NewtonRsCameraMeasurement¶

Newton-based rolling shutter projection

Projects a landmark into a rolling shutter camera using the Newton method. This method should produce measurements such that the projected image row is consistent with the projection time. However, this is not a guarantee, so if you really need to make sure that the rolling shutter prjection time constraint is fulfilled, we recommend to at least check the result before accepting it.

class kontiki.measurements.LiftingRsCameraMeasurement¶

Rolling shutter projection by time optimization

Projects a landmark into a rolling shutter camera by adding the projection time as an additional parameter to the optimization problem.

In general, this does not fulfill the rolling shutter projection time constraint exactly.

vt¶: The row projection time (to be optimized)

IMU measurements¶

The IMU measurements share the following base class

class kontiki.measurements.ImuMeasurement(imu, t, x, weight=1)¶

A measurement x (3D) at time t using the specified sensors.Imu instance. The scalar weight is applied to the residual, before being squared.

imu¶: The sensors.Imu instance used.

t¶: The timestamp of the measurement

weight¶: The (scalar) residual weight

IMU measurement classes¶

class kontiki.measurements.AccelerometerMeasurement¶

a¶: Measured acceleration

class kontiki.measurements.GyroscopeMeasurement¶

w¶: Measured angular velocity