ICRA 2020 - Determining and Improving the Localization Accuracy of AprilTag Detection

Authors: Jan Kallwies, Bianca Forkel and Hans-Joachim Wuensche

Dataset is available for download here!

A ROS package with our proposed refinement and filtering of parlty occluded tags can be found on GitHub: https://github.com/UniBwTAS/apriltags_tas

Please click on the example images to get the images in vector graphics format (PDF) including the ground truth position.

The following example images show that tags without perspective distortion can be localized more accurately. See also Fig. 6b in our paper.

Image / Tag-ID	GroundTruth	AprilTag3	AprilTagsCpp	ArUco	ArUcoOpenCV
circle/0005.png Tag #21
circle/0035.png Tag #21
circle/0090.png Tag #21

Please click on the example images to get the images in vector graphics format (PDF) including the ground truth position.

The following example images show the significantly higher error in the tag localization of AprilTag 3 and AprilTags C++ for straight tags. See also Fig. 7 in our paper.

Image / Tag-ID	GroundTruth	AprilTag3	AprilTagsCpp	ArUco	ArUcoOpenCV
roll_camera/0177.png Tag #21
roll_camera/0180.png Tag #21

The following example images show the significant change of the ArUco tag localization resulting from just a slight rotation of the tag of 4 degree. See also Fig. 7 in our paper.

Image / Tag-ID	GroundTruth	AprilTag3	AprilTagsCpp	ArUco	ArUcoOpenCV
roll_camera/0201.png Tag #21
roll_camera/0205.png Tag #21