OpenCV  4.8.0-dev
Open Source Computer Vision
Aruco module FAQ

Prev Tutorial: Calibration with ArUco and ChArUco

This is a compilation of questions that can be useful for those that want to use the aruco module.

In this case, you only need single ArUco markers. You can place one or several markers with different ids in each of the object you want to identify.

The aruco module is based on the original ArUco library. A full description of the detection process can be found in:

S. Garrido-Jurado, R. Muñoz-Salinas, F. J. Madrid-Cuevas, and M. J. Marín-Jiménez. 2014. "Automatic generation and detection of highly reliable fiducial markers under occlusion". Pattern Recogn. 47, 6 (June 2014), 2280-2292. DOI=10.1016/j.patcog.2014.01.005

There can be many factors that avoid the correct detection of markers. You probably need to adjust some of the parameters in the DetectorParameters object. The first thing you can do is checking if your markers are returned as rejected candidates by the detectMarkers() function. Depending on this, you should try to modify different parameters.

If you are using a ArUco board, you can also try the refineDetectedMarkers() function. If you are using big markers (400x400 pixels and more), try increasing adaptiveThreshWinSizeMax value. Also avoid narrow borders (5% or less of the marker perimeter, adjusted by minMarkerDistanceRate) around markers.

Using a board of markers you can obtain the camera pose from a set of markers, instead of a single one. This way, the detection is able to handle occlusion of partial views of the Board, since only one marker is necessary to obtain the pose.

Furthermore, as in most cases you are using more corners for pose estimation, it will be more accurate than using a single marker.

The main drawback is that a Board is not as versatile as a single marker.

ChArUco boards combines chessboards with ArUco boards. Thanks to this, the corners provided by ChArUco boards are more accurate than those provided by ArUco Boards (or single markers).

The main drawback is that ChArUco boards are not as versatile as ArUco board. For instance, a ChArUco board is a planar board with a specific marker layout while the ArUco boards can have any layout, even in 3d. Furthermore, the markers in the ChArUco board are usually smaller and more difficult to detect.

No. The main goal of ChArUco boards is provide high accurate corners for pose estimation or camera calibration.

No, the marker corners in a ArUco board can be placed anywhere in its 3d coordinate system.

Yes, all the markers in a ChArUco board need to be in the same plane and their layout is fixed by the chessboard shape.

The GridBoard class is a specific type of board that inherits from Board class. A GridBoard object is a board whose markers are placed in the same plane and in a grid layout.

Diamond markers are very similar to a ChArUco board of 3x3 squares. However, contrary to ChArUco boards, the detection of diamonds is based on the relative position of the markers. They are useful when you want to provide a conceptual meaning to any (or all) of the markers in the diamond. An example is using one of the marker to provide the diamond scale.

Yes, the detection of single markers is a basic tool in the aruco module. It is done using the detectMarkers() function. The rest of functionalities receives a list of detected markers from this function.

Yes, the aruco module provides functionalities to calibrate the camera using both, ArUco boards and ChArUco boards.

It is highly recommended the calibration using ChArUco board due to the high accuracy.

In general, it is easier to use one of the predefined dictionaries. However, if you need a bigger dictionary (in terms of number of markers or number of bits) you should generate your own dictionary. Dictionary generation is also useful if you want to maximize the inter-marker distance to achieve a better error correction during the identification step.

Dictionary generation should only be done once at the beginning of your application and it should take some seconds. If you are generating the dictionary on each iteration of your detection loop, you are doing it wrong.

Furthermore, it is recommendable to save the dictionary to a file with cv::aruco::Dictionary::writeDictionary() and read it with cv::aruco::Dictionary::readDictionary() on every execution, so you don't need to generate it.

Yes, one of the predefined dictionary is DICT_ARUCO_ORIGINAL, which detects the marker of the original ArUco library with the same identifiers.

Not directly, you will need to adapt the information of the ArUco file to the aruco module Board format.

Probably yes, however you will need to port the dictionary of the original library to the aruco module format.

If you are using one of the predefined dictionaries, it is not necessary. Otherwise, it is recommendable that you save it to file.

If you are using a GridBoard or a ChArUco board you only need to store the board measurements that are provided to the GridBoard::create() function or in or ChArUco constructor. If you manually modify the marker ids of the boards, or if you use a different type of board, you should save your board object to file.

You can use cv::aruco::Dictionary::writeDictionary() and cv::aruco::Dictionary::readDictionary() for cv::aruco::Dictionary. The data member of board classes are public and can be easily stored.

To do so, you will need to use an external rendering engine library, such as OpenGL. The aruco module only provides the functionality to obtain the camera pose, i.e. the rotation and traslation vectors, which is necessary to create the augmented reality effect. However, you will need to adapt the rotation and traslation vectors from the OpenCV format to the format accepted by your 3d rendering library. The original ArUco library contains examples of how to do it for OpenGL and Ogre3D.

You can cite the original ArUco library:

S. Garrido-Jurado, R. Muñoz-Salinas, F. J. Madrid-Cuevas, and M. J. Marín-Jiménez. 2014. "Automatic generation and detection of highly reliable fiducial markers under occlusion". Pattern Recogn. 47, 6 (June 2014), 2280-2292. DOI=10.1016/j.patcog.2014.01.005

It is important to remark that the estimation of the pose using only 4 coplanar points is subject to ambiguity. In general, the ambiguity can be solved, if the camera is near to the marker. However, as the marker becomes small, the errors in the corner estimation grows and ambiguity comes as a problem. Try increasing the size of the marker you're using, and you can also try non-symmetrical (aruco_dict_utils.cpp) markers to avoid collisions. Use multiple markers (ArUco/ChArUco/Diamonds boards) and pose estimation with solvePnP() with the SOLVEPNP_IPPE_SQUARE option. More in this issue.