Ejemplo n.º 1
0
 def test_get_positions(self):
     # create built-in array of coordinates (lat-lon)
     coordinates = np.array([(44.484372, 11.355899), (44.484171, 11.355872),
                             (44.483912, 11.355410),
                             (44.483966, 11.355011)])
     coordinates = np.deg2rad(coordinates)
     # altitude is zero
     altitudes = np.zeros(4)
     # create expected distance (values are taken from a website)
     expected_distance = np.array([0, 22.45, 64.20, 83.67])
     # call method to test
     positions, _ = get_positions(coordinates.T, altitudes)
     # calculate distance from (0,0) with norm
     distance = np.linalg.norm(positions, axis=0)
     # ground truth is calculated online with a website that use haversine distance, while get_position doesn't
     # so limit equality to 2 decimal digit. it would be enough for small distances (gnss is 40hz)
     np.testing.assert_array_almost_equal(expected_distance,
                                          distance,
                                          decimal=2)

import matplotlib.pyplot as plt

if __name__ == '__main__':
    window_size = 20
    path = '../tests/test_fixtures/parking.tsv'

    # currently default format is unmodified fullinertial but other formats are / will be supported
    times, coordinates, altitudes, gps_speed, heading, accelerations, angular_velocities = parse_input(path, [
        InputType.UNMOD_FULLINERTIAL])

    converts_measurement_units(accelerations, angular_velocities, gps_speed, coordinates)

    # get positions from GNSS data
    gnss_positions, headings = get_positions(coordinates, altitudes)

    # reduce accelerations disturbance
    times, accelerations = reduce_disturbance(times, accelerations, window_size)
    # reduce angular velocities disturbance
    _, angular_velocities = reduce_disturbance(times, angular_velocities, window_size)
    # truncate other array to match length of acc, thetas, times array
    gnss_positions = gnss_positions[:, round(window_size / 2):-round(window_size / 2)]

    # with "final" times now get velocities and
    real_velocities = get_velocities(times, gnss_positions)
    # scalar speed from GNSS position (better than from dataset because avoids Kalmar filter)
    real_speeds = np.linalg.norm(real_velocities, axis=0)

    # get time windows where vehicle is stationary
    stationary_times = get_stationary_times(gps_speed)
Ejemplo n.º 3
0
def get_trajectory_from_path(path):
    """
    parse input file from path, clean data and integrate positions

    :param path: string input file
    :return: 3 numpy array: 3xn position, 1xn times, 4xn angular position as quaternions
    """

    window_size = 20

    # currently default format is unmodified fullinertial but other formats are / will be supported
    times, coordinates, altitudes, gps_speed, heading, accelerations, angular_velocities = parse_input(
        path, [InputType.UNMOD_FULLINERTIAL])

    converts_measurement_units(accelerations, angular_velocities, gps_speed,
                               coordinates, heading)

    # get positions from GNSS data
    gnss_positions, headings_2 = get_positions(coordinates, altitudes)

    # reduce accelerations disturbance
    times, accelerations = reduce_disturbance(times, accelerations,
                                              window_size)
    # reduce angular velocities disturbance
    _, angular_velocities = reduce_disturbance(times, angular_velocities,
                                               window_size)
    # truncate other array to match length of acc, thetas, times array
    gnss_positions = gnss_positions[:,
                                    round(window_size /
                                          2):-round(window_size / 2)]

    # with "final" times now get velocities and
    real_velocities = get_velocities(times, gnss_positions)
    # scalar speed from GNSS position (better than from dataset because avoids Kalmar filter)
    real_speeds = np.linalg.norm(real_velocities, axis=0)

    # get time windows where vehicle is stationary
    stationary_times = get_stationary_times(gps_speed)

    # clear gyroscope drift
    angular_velocities = clear_gyro_drift(angular_velocities, stationary_times)
    # set times start to 0
    normalize_timestamp(times)

    # correct z-axis alignment
    accelerations, angular_velocities = correct_z_orientation(
        accelerations, angular_velocities, stationary_times)

    # remove g
    accelerations[2] -= accelerations[
        2, stationary_times[0][0]:stationary_times[0][-1]].mean()

    # correct alignment in xy plane
    #accelerations = correct_xy_orientation(accelerations, angular_velocities)

    motion_time = get_first_motion_time(stationary_times, gnss_positions)
    initial_angular_position = get_initial_angular_position(
        gnss_positions, motion_time)

    # convert to laboratory frame of reference
    accelerations, angular_positions = rotate_accelerations(
        times, accelerations, angular_velocities, heading,
        initial_angular_position)

    # rotate to align y to north, x to east
    accelerations = align_to_world(gnss_positions, accelerations, motion_time)
    # angular position doesn't need to be aligned to world if starting angular position is already aligned and following
    # angular positions are calculated from that

    initial_speed = np.array([[gps_speed[0]], [0], [0]])
    # integrate acceleration with gss velocities correction
    correct_velocities = cumulative_integrate(times,
                                              accelerations,
                                              initial_speed,
                                              adjust_data=real_velocities,
                                              adjust_frequency=1)

    if sign_inversion_is_necessary(correct_velocities):
        accelerations *= -1
        correct_velocities *= -1

    correct_position = cumulative_integrate(times,
                                            correct_velocities,
                                            adjust_data=gnss_positions,
                                            adjust_frequency=1)

    return correct_position, times, angular_positions