apriltag_pose_estimation

apriltag_pose_estimation is a Python library for pose estimation using AprilTags. It abstracts all the OpenCV and AprilTag library code with a Pythonic interface designed to make tuning parameters easy. It hides much of the matrices and other technical aspects to make things as easy as possible for beginners.

Installation

This library currently isn't available on PyPI. However, you can perform a local installation by running:

pip install ./apriltag_pose_estimation

Add -e if you plan to modify the library.

Basic usage

Specify an AprilTag field layout in JSON:

{
  "fiducials": [
    {
      "id": 0,
      "rotation_vector": [
        -1.2091995761561456,
        1.2091995761561452,
        -1.2091995761561458
      ],
      "translation_vector": [
        0,
        0.105,
        0.56
      ]
    }
  ],
  "tag_size": 0.080,
  "tag_family": "tagStandard41h12"
}

Load the field:

from apriltag_pose_estimation.core import load_field

with open('examples/mars_testfield.json', mode='r') as f:
    field = load_field(f)

Specify the parameters of the camera being used:

from apriltag_pose_estimation.core import CameraParameters

camera_params = CameraParameters(fx=1329.143348,
                                 fy=1326.537785,
                                 cx=945.392392,
                                 cy=521.144703,
                                 k1=-0.348650,
                                 k2=0.098710,
                                 p1=-0.000157,
                                 p2=-0.001851,
                                 k3=0.000000)

Create a pose estimator:

from apriltag_pose_estimation.localization import CameraLocalizer
from apriltag_pose_estimation.localization.strategies import (MultiTagPnPStrategy,
                                                              LowestAmbiguityStrategy)

estimator = CameraLocalizer(
    strategy=MultiTagPnPStrategy(fallback_strategy=LowestAmbiguityStrategy()),
    field=field,
    camera_params=camera_params,
    nthreads=2,
    quad_sigma=0,
    refine_edges=1,
    decode_sharpening=0.25
)

Estimate a pose from a grayscale image:

image = # ...image...
result = estimator.estimate_pose(image)
print(result.estimated_pose)

Rendering camera position

It can be useful for troubleshooting to see where the camera is in space. For this reason, we provide a render subpackage which renders the current camera position in a Qt window.

Create a CameraPoseDisplay and initialize with an AprilTagField:

from apriltag_pose_estimation.localization.render import CameraPoseDisplay

display = CameraPoseDisplay(field)

When a new camera pose is found, update the display:

display.update(estimated_pose)

You may need to execute the application, which can be done with display.exec_application(). In this case, you'll need to update the display asynchronously, such as with a QTimer.