Overview¶

Kontiki was created to estimate a continuous-time trajectory, and any auxiliary parameters, given a set of measurements.

To use kontiki you need to understand, and create, objects of the following types:

Trajectories: A continuous-time trajectory that you want to estimate
Measurements: The optimizer tries to estimate the trajectory (and other optimizable entities) such that the measurement errors are minimized.
Sensors: A sensor, such as a camera or an IMU, models how a measurement was created. Some sensors have parameters that can be estimated if you want to.
Data objects: Some measurements are connected to more complex objects, which might in turn be available for estimation. A prime example are Camera measurements, where the associated Landmark is an object that we want to estimate as well.

The trajectory, and set of measurements, are then added to a TrajectoryEstimator instance that can then minimize the measurement errors.

Example visual-inertial structure from motion estimation¶

import numpy as np
import itertools
from kontiki import trajectories, sensors, TrajectoryEstimator
import kontiki.measurements as M

views, landmarks = load_my_sfm_data()

trajectory = trajectories.UniformSE3SplineTrajectory()
trajectory.extend_to(some_max_time, np.eye(4))

imu = sensors.BasicImu()

camera = sensors.PinholeCamera(720, 1280, 0.03, my_camera_matrix)

gyro_meas = [M.GyroscopeMeasurement(imu, t, x)
             for t, x in my_gyro_data]
acc_meas = [M.AccelerometerMeasurement(imu, t, x)
            for t, x in my_acc_data]
im_meas = [M.StaticRsCameraMeasurement(camera, obs)
           for v in view for obs in v.observations]

estimator = TrajectoryEstimator(trajectory)
for m in itertools.chain(gyro_meas, acc_meas, im_meas):
    estimator.add_measurement(m)

summary = estimator.solve()