Fiducial

Fiducial detection and recognition differs from that of generic object recognition as some assumptions are made on the type of object being detected. For example, a black and white fiducial can be detected without needing color, and can be easily separated from the background due to its high contrast nature. Fiducials are also typically 2D planar objects that have very distinct corners or shapes. Identifying these traits in the environment helps to reduce the "clutter" of other objects and ensures better identification reliability.

Thanks to the planar nature of fiducials they can be placed on the floor, ceiling or walls and still be detected correctly without any recalibration of the camera. By knowing the basic shape of a fiducial machine vision techniques can account for the perspective and affine distortion (i.e. a fiducial rotated in X and Y plane) to a considerable degree while still maintaining recognition. Is is for this reason, along with their barcode like nature, that fiducials are frequently used for navigation and localization.

There are 6 parameters that localize a fiducial in space. Luckily, due to the shape and planar nature of the fiducial these 6 parameters can be determined very quickly and allow the processing to run at very near camera speed. This has an advantage on moving robots that need to process their environment before it changes.

Once you know the identity of a fiducial you can know approximately where you are. Having the fiducial in sight helps the robot to know where it is with respect to the fiducial and can use the fiducial as a target for navigation based on scale, rotation and translation.

The Fiducial module provides a way to easily recognize fiducials in your environment. We have included 15 test fiducials in the Fiducials folder along with the RoboRealm download that you can print out to experiment with. If you don't have a printer just bring them up on your screen and point your camera towards it!

Note that these are graphical files that can be modified (or more created) as needed. Also note the one poster like fiducial. The only requirement for the fiducial is that it be surrounded by a white surface with an inner black square and contain inner elements that can distinguish it from other Fiducials. If you have posters that satisfy this requirement then you already have fiducials on your walls!

Interface

Instructions

1. Train - Press the train button and enter the full path to the Fiducials folder that you downloaded from RoboRealm. This will typically be c:\Program Files\RoboRealm\Fiducials\. Press the START button to start training. After it is complete press the ok button.

2. Display - You should know notice the fiducials being recognized in the main RoboRealm window. Matches are displayed as green whilst candidate regions that do not match are in purple. Note that the 'name' of a fiducial is the name of the image filename without the extension. So you can change the names by changing the filenames and re-training.

3. Attributes - By default the Filename and Confidence are displayed. You can add or remove information from being displayed by changing the checkboxes in the Display area. Note if too much information is displayed you might want to change the Font size to a smaller size.

• Filename - the name of the fiducial.
• Confidence - how strongly the module believes this fiducial matches one in the database folder.
• Coordinates - the location of the matched fiducial.
• X Translation - the X location with respect to the center of the screen of the fiducial.
• Y Translation - the Y location with respect to the center of the screen of the fiducial.
• Size - the size of the fiducial relative to that in the database.
• X Rotation - the rotation in the X axis of the fiducial (pan)
• Y Rotation - the rotation in the Y axis of the fiducial (tilt)
• Orientation - the Z rotation or angle of the fiducial (spin)

4. Box Color - You can change the match and test box color by changing the Matched Box Color and Test Box Color respectively.

5. Filter Confidence - Many candidate squares do pass the basic fiducial requirements but do not match with a known fiducial very well. You can remove those unwanted matches by increasing the Min Confidence. This removes bad matches from being made.

6. Filter Size - Each recognized fiducial has a scale associated with it that can be used to remove larger or smaller matches.

7. Filter Orientation - by default fiducials are recognized in any orientation. You can use the Filter Orientation to remove those fiducials outside of your desired orientation. For example, 15, 345 would remove any fiducial +=15 degrees from that what is in the database.

8. Filter Threshold - you can decide on how many fiducials you want to work with. Specifying a number in the Filter Threshold will display the best X fiducials found in the image.

9. Stabilization - When moving the camera motion blurring can cause incorrect low confidence matches to appear. By increasing the Occurrence Threshold you are requiring a fiducial to appear and continue to appear for a set number of frames before being reported as a match. This essentially removes the "flickering" of low matched fiducials that can cause noise in the results. Note that if you are working with static images you will need to set this to zero!

10. Perspective - Due to the depth of field of the camera perspective will be apparent in the fiducial square. You need to specify what this depth of field is for your camera in order for the fiducial's angles to be extracted correctly. The manufacturer of your camera will normally know this value. If you do not have access to this value you can empirically determine this by placing a fiducial at a known angle to the camera (for example 45 deg X rotation) and adjusting the depth of field FL value up or down until the X rotation correctly reads 45.

11. Variables - To use the results of the fiducial module in your own application or in other RoboRealm modules you can select that the fiducial array be created. This array has all the collected information about a detect fiducial. See below for the array format.

12. Display Results - By default No Change is selected which does not modify the current image in any way other than annotating the fiducials with the above colors. If you want to use the results in other modules you can select to see

• Colored - displays only the fiducials with their original pixel colors
• Inverted - displays everything else but the fiducial (which is blacked out).
• White Mask - white where the fiducials are, black everywhere else
• Black Mask - black where the fiducials are, white everywhere else

Example

Source	Fiducial

Variables

FIDUCIALS - contains an array of detected fiducials.

The array contains blocks of 17 values per detected fiducial. The
elements are as follows:

The FIDUCIALS array is composed of 17 numbers as follows:
Offset      Contents
0           Match Confidence 0-100
1           Point 1 X coordinate
2           Point 1 Y coordinate
3           Point 2 X coordinate
4           Point 2 Y coordinate
5           Point 3 X coordinate
6           Point 3 Y coordinate
7           Point 4 X coordinate
8           Point 4 Y coordinate
9           Translation in X
10          Translation in Y
11          Scale 0-100
12          Rotation in X (radians)
13          Rotation in Y (radians)
14          Orientation (Rotation in Z, radians)
15          Path start index
16          Length of path

For example, to print out all found fiducials use the VBScript module
and the following script.

fiducials = GetArrayVariable("FIDUCIALS")
names = GetStrVariable("FIDUCIALS_PATH")

if isArray(fiducials) then
  if ubound(fiducials) > 0 then
    for i=0 to ubound(fiducials)-1 step 17
      nstart = fiducials(i+15)
      nend = fiducials(i+16)
      write "Conf: " & fiducials(i) & "%  Path: " & _
        mid(names, nstart, nend) & vbCRLF
    next
  end if
end if

For ease of access the most confident fiducials' information is
also available in the following variables

FIDUCIAL_CONFIDENCE - the confidence of the best matched fiducial
FIDUCIAL_NAME - the name of the matched fiducial
FIDUCIAL_FILENAME - the filename of the fiducial
FIDUCIAL_FOLDER_X - the folder sequence that the fiducial is in
  (replace X with 1, 2, etc.)
FIDUCIAL_SIZE - the size of the best matched fiducial
FIDUCIAL_X_COORD - the x location relative to center screen of
  the best matched fiducial
FIDUCIAL_Y_COORD - the y location relative to center screen of
  the best matched fiducial
FIDUCIAL_ORIENTATION - the orientation (spin) of the best fiducial
FIDUCIAL_X_ROTATION - the x rotation (pan) of the best fiducial
FIDUCIAL_Y_ROTATION - the y rotation (tilt) of the best fiducial

See Also

For more information