Understanding AprilTag Detection Values
Last Updated: 7/05/2023
When an AprilTag is detected by the new SDK vision processing system, the core code returns a collection of raw data that is often not easily interpreted. However, the data can be further transformed into a familiar frame of reference to make it more easily utilized.
In the FIRST Tech Challenge SDK, the AprilTag API will present the Team OpMode with a collection of translation and rotation values, called ftcPose, that represent the Tag’s position in 3D space.
To understand how to interpret these values, it’s easier to consider a simpler 2D scenario where the vertical component is ignored. This is what is described below.
Figure 1 below represents one possible 2D scenario.
This diagram looks down on the Camera and AprilTag from above. The camera’s “forward” direction is identified by a dashed line drawn straight out from the camera.
The AprilTag image is shown in the upper left of the figure. The tag is located 100 units forward of the camera and 36.4 units to the left (measured at right angles to the forward view).
The AprilTag is also rotated 5 degrees counterclockwise from a normal “face on” orientation.
Now that we have a clear understanding of one possible detection scenario, we can look at the meaning of the various values returned as ftcPose by the SDK.
Figure 2 shows the measured values associated with the camera/target scenario shown in Figure 1.
Since this is a simple 2D diagram, the vertical “Z” (up) axis is being ignored, so it is not shown here.
The green X axis value represents the sideways offset to the tag. Note that this value is negative (to the left of the camera center).
The red Y axis value represents the forward distance to the Tag.
The cyan Yaw value represents the rotation of the tag around the Z axis. A Counter-Clockwise rotation is considered positive. Note that a Yaw value of zero means that the tag image is parallel to the face of the camera.
Fun Fact: If the camera is pointing forward, the X, Y & Z axes are consistent with the Robot Coordinate system.
Three additional parameters are derived from the X and Y axis values, these are Range (which is the direct distance to the center of the target), Bearing (which is how many degrees the camera must turn to point directly at the target) and Elevation (which is how many degrees the camera must tilt UP to center on the tag). Note that Target Bearing has the same positive counterclockwise orientation.
Investigating some real data
To illustrate this process, consider some real-world tags. The data that follows is from a pair of tags printed on a card. The left-most tags has an identical setup as described above. In Figure 3 the protractor origin is positioned directly in front of the camera, at a distance of 25 inches. Both tags are to the left of the camera centerline, and both are rotated +5 degrees. The left tag is 6.6” from the centerline, and the right tag is 1.5” from the centerline. The camera is located 6” inches above the center of the targets, looking out horizontally (parallel to the ground).
The AprilTag video preview image from the Camera Stream preview is shown below. The left tag has an ID of 0 and the right tag has an ID of 1. This video is being captured by a Logitech C920 Pro HD webcam, running at 648x480 resolution. In this mode the camera has Field-Of-View (FOV) of 60 degrees. The physical tags are 3.4” square.
Notice that both tags are in the bottom-left corner of the image. The center of the image corresponds to the location the camera is pointed at, which is centered on the protractor and directly above the top of the tags.
Based on this setup, let’s review the data returned by the “ConceptAprilTag.java” sample OpMode.
Since the creation of this document, the tags used in the ConceptAprilTag.java sample have changed. Therefore, in order to reproduce this example the appropriate tags will need to be used instead of Tag0 and Tag1.
The values for the two AprilTags are listed as “ID0 Nemo”, and “ID1 Jonah”. These are the names assigned when adding the tags to the Tag Library.
The OpMode displays values that correspond to those parameters shown in Figure 2. The XYZ line shows the three axes translation values (X, Y & Z) in inches. The PRY line shows the corresponding rotations (Pitch, Roll & Yaw) around those axes, in degrees. The RBE line shows the target Range (in inches), Bearing, and Elevation (in degrees). The angle of Elevation results from the height difference between the camera and the Tag.
Several items to observe:
Both Y values are about 25”, but the Y value for Tag 1 is slightly larger because it is behind the protractor base line.
The X values for Tag 0 and 1 correspond to the offset distances described earlier (-6.6” and -1.5”)
Both tags show a Yaw of approximately 5 Deg, although this can vary 1-2 degrees depending on other orientation factors.
The Range to both targets are almost equal but the Bearing of Tag 0 is much greater due to its displacement to the left.
Both targets show the same negative Z value of -5.7, which is consistent with them being centered about 6” below the height of the camera.
Each tag also has an “Elevation” of about -12.6 degrees, which is a downward viewing angle to the center of each tag.
Ways to use this data
There are several ways the AprilTag position data can be used, but here are two basic ways.
Pointing towards a target (Tank Drive).
If an AprilTag is being used to mark the location of a target that you need to shoot towards, then the two main properties of interest are Tag Range and Tag Bearing. The Tag Bearing is an indication of how many degrees you would need to turn to point directly at the tag, and the Tag Range is an indication of how far you would need to shoot. Even with a simple differential (tank) drive, these two parameters would enable you to turn towards the target and drive to the correct range (or adjust your shooting power based on the range).
A simple proportional controller could take the Tag Bearing, multiply it by a suitable gain and then use it in place of the turning joystick to turn the robot towards the target. Likewise, you could subtract the desired shooting range from the current Tag Range and use the result to control the robot’s forward speed.
Note that this approach does not guarantee that you are squared up to the front of the target, merely that you are pointing towards it. To get squarely aligned, you need to consider the Yaw angle as shown in the next approach.
See SDK Sample: RobotAutoDriveToAprilTagTank.java for more info.
Approaching a target squarely (Omni Drive).
If an AprilTag is being used to mark the location of something that must be approached squarely from the front, then it’s important to consider the Tag Yaw value. This is a direct indication of how far off (in degrees) the camera is from the tag image’s centerline. This is related to, but not the same as the Tag Bearing. So, all three parameters (Range, Bearing & Yaw) must be used to approach the target and end up directly in front of it.
Reaching a certain distance directly in front of the target can be easily performed by a robot with a holonomic (Omnidirectional) drive, because strafing can be used for direct sideways motion. A three-pronged approach can be used. 1) The Target bearing can be used to turn the robot towards the target (as described above). 2) The Target Yaw can be used to strafe sideways, thereby rotating around the target to get directly in front of it. 3) The target range can be used to drive forward or backward to obtain the correct standoff distance.
Each of the three axis motions could be controlled by a simple proportional control loop, where turning towards the tag is given the highest gain (priority), followed by strafing sideways, followed by approaching the tag.
See SDK Sample: RobotAutoDriveToAprilTagOmni.java for more info.