You launch the tracker, open the result topic in the image viewer, and there is the marker: outlined, labeled, detected. Then you regenerate the tf tree, and no aruco_marker frame appears anywhere in frames.pdf. This is the set-up where detection works and publishing does not — usually a depth camera or an area scan camera feeding an aruco_ros tracker ahead of hand-eye calibration. Four causes cover nearly every case, and they separate cleanly: nothing broadcasts the frame, the remaps point at topics that do not exist, the marker parameters do not match the printout, or the rectification flag does not match the stream. Work them in the order below and finish with the tf_echo check that proves the frame.
What the overlay proves, and what it does not
The overlay drawn on /aruco_single/result proves three things: images reach the tracker, the dictionary matches the printout, and the marker ID matches. That is all detection. The transform is a separate output, broadcast by a node configured to publish it, and it depends on inputs the overlay never shows — the two remapped camera topics and the parameters describing the marker.
That separation is why the frame can vanish while detection carries on. The node drawing the overlay and the node broadcasting aruco_marker can be two processes with two different remap lines. When the publisher's inputs are wrong, the console stays quiet: no error, no warning. The first visible sign is the missing frame in the tree.
Start by learning the exact topic names your driver publishes. The procedure for viewing the image topics a depth camera driver runs prints them in one pass, and you will paste those names into the remaps.
The four causes, and the test that separates each
| Cause | Symptom | Test | Fix |
|---|---|---|---|
| Nothing broadcasts the frame | Overlay draws, no aruco_marker in the tree | Read which nodes your launch file starts | Add the marker publisher node |
| Remaps name topics that do not exist | Overlay draws on the tracker, publisher silent | Paste each remap target into the image viewer | Point /camera_info and /image at the driver's color stream |
| marker_id or dictionary mismatch | No overlay either | Read the printed ID beside the marker | Set the matching ID, print the original dictionary |
| marker_size or image_is_rectified wrong | Frame appears, numbers fail the tape test | Compare tf_echo with a measured stand-off | Type the caliper value in metres, match the flag to the stream |
The first two keep the frame out of the tree; the last two publish a frame you cannot trust. Most set-ups arriving at this symptom sit in row one or row two.
Fix 1: run the node that broadcasts the frame
The tracker you launched draws the overlay; the aruco_marker frame is broadcast by the marker publisher. Put both in one launch file, remapped to the same camera:
<launch>
<node name="aruco_tracker" pkg="aruco_ros" type="single">
<remap from="/camera_info" to="/camera/rgb/camera_info" />
<remap from="/image" to="/camera/rgb/image_rect_color" />
<param name="marker_size" value="0.1" />
<param name="marker_id" value="12" />
<param name="image_is_rectified" value="True" />
</node>
<node pkg="aruco_ros" type="marker_publisher" name="aruco_marker_publisher">
<remap from="/camera_info" to="/camera/rgb/camera_info" />
<remap from="/image" to="/camera/rgb/image_rect_color" />
<param name="marker_size" value="0.1" />
<param name="image_is_rectified" value="True" />
</node>
</launch>
Only the two remap targets are yours to change at first; the rest is marker data you already hold. The values shown are a working example: a 100 mm marker, ID 12, on the rectified color stream. Restart the launch, regenerate the tree, and aruco_marker should appear. If it still does not, the remaps are next.
Fix 2: point both remaps at topics that exist
Left unremapped, the nodes subscribe to /camera_info and /image at the top of the topic space, and most drivers publish under their own namespace — so the nodes wait forever. A node subscribed to a topic nobody publishes receives nothing and publishes nothing.
Test each remap target by feeding the exact string to the image viewer:
rosrun image_view image_view image:=/camera/rgb/image_rect_color
A blank window means the string is wrong: check the namespace, the spelling, and that the driver is running. The two topics are a pair, not two independent lines. The calibration carried by camera_info must belong to the exact image stream you remap — pairing a color stream with another camera's calibration produces a frame that sits in the tree and is wrong in every number. Compare the frame_id and the image dimensions in the camera_info message against the stream you remapped; they must describe that stream and nothing else.
Fix 3: make the ID, the dictionary and the size match the printout
marker_id 12 means the tracker follows marker 12 and nothing else; if your overlay is blank as well as the frame missing, read the printed ID first. Generate the printout with the original ArUco dictionary — the other dictionaries a generator page offers print similar-looking grids this tracker will not follow.
marker_size is the side of the marker square in metres: 0.1 for a 100 mm marker. Measure the print with calipers instead of trusting the printer, as measuring the marker before typing its size shows, and watch the decimal separator — a locale that writes commas can silently change 0.1, and the comma-decimal marker size fix locates it.
A wrong ID keeps the frame away entirely. A wrong size publishes a frame whose every distance is scaled, which is worse because nothing looks broken.
Worked example: a 120 mm print typed in as 0.1
Inputs: the printout measures 120 mm across the marker square (calipers); the launch says marker_size 0.1, which is 100 mm; the camera stands 0.60 m from the marker (tape measure on the bench).
Rule: the pose is solved from the square's size in the image, so every coordinate scales with declared size over true size.
Arithmetic: scale = 0.100 m / 0.120 m = 0.833. Published distance along the optical axis = 0.833 × 0.60 m = 0.50 m. The reading is 0.10 m short, and 0.10 m / 0.60 m = 17 %. X and Y shrink by the same factor, so the tool stops 17 % short along every axis.
Decision: type the caliper value — here 0.12 — and re-run the check below. Any residual difference between calipers and parameter is that same percentage error on every distance the frame will ever give. If the reading then matches the tape at one stand-off, it matches at all of them.
Fix 4: set image_is_rectified to the stream you feed
image_is_rectified set to True states that the input images are already undistorted, which is exactly what the rectified color stream is. Keep the flag and change the stream: remap to the rectified topic and leave True. The other combination — a raw stream with the flag still True — publishes a frame with lens distortion left inside it, small near the image centre and growing toward the corners. The flag describes the stream; it does not repair one.
Verify with tf_echo before anything moves
- Bring the marker into view and confirm the overlay still draws.
- Run tf_echo from your camera's optical frame down to the marker:
rosrun tf tf_echo <camera_optical_frame> aruco_marker
The parent name is the frame your driver names in its calibration message — the same frame camera_info carries.
- Compare the translation along the optical axis with the tape-measured stand-off. Expect agreement within a percent or two; a constant percentage offset points straight at marker_size, and a Z that misbehaves while X and Y hold up has its own fix in the ArUco Z axis distance error.
- Regenerate the tf tree; aruco_marker must now sit between the camera frame and your calibration set-up. From here, publishing the marker pose as a tf frame for hand-eye calibration takes over, and a calibration that still cannot find its target needs the hand-eye target frame checks.
Once the frame is live, aligning a robot gripper with a marker coordinate frame is the natural next step.
The camera that feeds the tracker
Any camera whose driver publishes a rectified color stream together with its calibration message can feed this tracker; the hardware choice comes down to motion. A global shutter keeps marker edges crisp while the camera or the part moves during exposure; a rolling shutter is equally suitable at a static calibration station. The area scan cameras finder covers both.
GigE, USB3, 10GigE and CoaXPress
1,287 models · from $149.00For a moving cell:
- Sensor format
- 1/1.7"
- Lens mount
- C-mount
- Interface
- GigE
- Resolution
- Up to 3 MP
The MV-CU020-90GC is a 2.5 MP GigE color camera with a global shutter and a 1/1.7" sensor. For a static station the MV-CS200-10UM is a 20 MP USB3.0 mono camera with a 1" sensor — the detector reads grayscale edges, not color, so mono gives up nothing.
- Sensor format
- 1"
- Lens mount
- C-mount
- Interface
- USB3
- Resolution
- 16-25 MP
If the calibration rig needs depth as well as the marker frame, the 3D sensors finder lists laser profilers and stereo 3D cameras.
Laser profilers and stereo 3D cameras
170 models · from $699.00Checklist: from missing frame to trusted frame
- The launch file starts a marker publisher, not only the tracker that draws the overlay — the overlay never proves a frame exists.
- Every remap target shows an image when pasted into the viewer unchanged, because a mistyped namespace fails silently.
- camera_info and the image come from the same camera, and the stream is the rectified color one.
- marker_size equals the caliper reading across the marker square, typed in metres.
- marker_id matches the printed ID, and the printout came from the original ArUco dictionary.
- image_is_rectified is True on a rectified stream.
- tf_echo matches the tape at the working stand-off before any robot motion consumes the frame.
When the frame is solid, the configurator started at the camera step sizes the rest of the chain around it. If the frame still resists after these four fixes, send your launch file and topic names through the contact page and we will read them with you.
Questions
Why does the marker show in the result image while the frame stays missing?
Because they are separate outputs. The overlay comes from the detector inside the tracker node and proves only that images arrive and the printout matches. The transform is broadcast by a publisher with its own remaps and parameters, and a node subscribed to topics nobody publishes fails without any error in the console, so the tf tree is where the gap first shows. Check which node your launch actually starts, then its two remap targets.
Should marker_size be the black square or the whole printout?
The side of the marker square itself, the black region, in metres; the white border around it is not counted. A marker that measures 100 mm across the black square takes 0.1, and a 120 mm print takes 0.12. Measure with calipers after printing: printers scale pages, and whatever error the print carries passes straight into every distance the frame publishes.
What happens if I feed the raw color stream with image_is_rectified still True?
The frame appears and the tree looks correct, but lens distortion stays inside the pose. The error is small near the image centre and grows toward the corners, so a marker dead-centre can read well while one at the edge is off. The fix is the stream, not the flag: remap to the rectified color topic and keep the flag True.
How do I prove the frame is right once it appears?
Check it against a tape measure at two stand-offs. Hold the marker at 0.40 m, then at 0.80 m, and compare the tf_echo translation along the optical axis with the tape each time. A constant percentage error at both distances means marker_size is wrong; an error that changes erratically between the two points at the camera calibration instead.
Can I print the marker from any generator page?
Use a generator that offers the original ArUco dictionary and pick that one; the other dictionaries on the same page produce similar-looking grids this tracker will not follow. Note the ID the generator assigns and put the same number in marker_id: a value of 12 means only marker 12 is tracked, whatever else is in view.
Do I need a color camera for ArUco tracking?
No. The detector works on grayscale edges, so a mono camera tracks markers just as well and is the usual choice when color adds nothing to the task. Color earns its place only when the same image must serve a color inspection, such as a presence check or label reading, beside the marker.

