You fill in the measured ArUco board dimensions in the calibration panel, select your colour image topic, and RViz freezes with Invalid target measured dimensions: marker_size 0,000000, marker_seperation 0,000000. The numbers you typed were not zero, so the panel has thrown them away. On a Linux box running a comma-decimal locale, the panel stops reading each value at the decimal comma and stores exactly 0 — and the error prints that comma back at you. This page shows you how to confirm the cause in one minute, fix it with LC_NUMERIC=en_US.UTF-8 (the case that exposed it ran Ubuntu 18.04 with ROS Melodic), and enter the dimensions so they store correctly the first time.
What the error message tells you
The message answers two questions at once, so read it as a diagnostic before touching any setting.
The commas prove the locale. A number printed as 0,000000 uses the decimal mark of the locale the process inherited from your desktop. If your terminal shows dots (0.000000), the locale is not your problem and the other causes below matter more.
The zeros prove a failed parse. The panel reads marker_size and marker_seperation when it builds the board target as the first image frame arrives, which is why the error fires at topic selection and not while you type. A parser expecting a dot reads 0,0502 up to the comma, keeps the 0, and discards the rest. Both fields end at exactly zero, validation rejects them, and the panel stops responding — kill the process and work from a fresh launch.
Likely causes, in order of probability
Work down the list; each cause has a one-step test.
-
Comma-decimal locale. Most likely when the error prints commas and zeros despite values you typed. Test: type
0.0502with a dot intomarker_sizeand re-select the topic. If the error disappears, the parse was the failure. - Values never committed. Some panels only store the fields when capture starts, so a field that reset to zero looks typed but is empty. Test: retype both values with dots after selecting the topics, then restart capture.
-
Empty camera info. A panel with no CameraInfo topic selected can behave as if nothing about the camera is known. Test: select the CameraInfo topic before the image topic. It is the right order anyway; but if the dimensions still print as
0,000000with commas, the camera is not the cause — return to cause 1.
Confirm the locale in one minute
In the shell you launch RViz from, run:
locale
Read the LC_NUMERIC line. Any language that writes decimals with a comma — German, French, Dutch and Belgian settings are the common ones — will reproduce the error. en_US.UTF-8 writes a dot.
The functional test is faster than reading settings: type 0.0502 with a dot and select the topic again. Dot entries parse even in a comma locale, because the comma is the one character the parser treats as the end of the number. A dot entry that stores while a comma entry stores zero confirms the diagnosis without changing anything.
The fix: run the panel under LC_NUMERIC=en_US.UTF-8
Export the numeric locale, then start RViz from the same shell:
export LC_NUMERIC=en_US.UTF-8
rviz
For a single launch, set it inline:
LC_NUMERIC=en_US.UTF-8 rviz
Confirm with locale: the LC_NUMERIC line should now read en_US.UTF-8. The export changes one locale category for the programs that shell starts — number formatting only. Your desktop language, dates and messages stay as they are, and closing the terminal ends the effect. LC_NUMERIC is the category of the C locale model that governs decimal marks and nothing else — the split is described in the locale section of the C library manual (man 3 locale), useful if you set the locale from a launch script.
Put the export in your launcher if you calibrate often, so every session starts with a dot decimal.
The fix covers any camera feeding the panel — a 3D sensor or an area scan camera — because the failure sits in the text fields, before any image is touched.
Worked example: a 50.2 mm marker under a comma locale
Inputs. Caliper readings from the printed board: marker side 50.2 mm, separation between adjacent marker origins 71.0 mm. Panel fields: marker_size and marker_seperation, in metres. Shell: LC_NUMERIC set to a comma-decimal language.
Rule. Millimetres convert to metres by dividing by 1000. A dot-expecting parser keeps characters up to the first comma and drops the rest.
Arithmetic:
50.2 mm ÷ 1000 = 0.0502 m71.0 mm ÷ 1000 = 0.0710 m- Typed as
0,0502: the parser keeps0and stops at the comma, so the stored value is0.000000 mand the panel printsmarker_size 0,000000. - Typed as
0.0502: the stored value is0.050200 m, six decimals — exactly what a healthy echo looks like.
Result. Comma entries leave both fields at exactly zero, matching the error; dot entries store the measured values.
Decision. Any error printing 0,000000 after a non-zero entry is a parse failure at the decimal mark: fix the locale or type dots, and never scale a number to compensate. Entering 50.2 instead of 0.0502 makes the board 50.2 ÷ 0.0502 = 1000 times too large, so a marker truly 0.75 m away solves at 0.75 × 1000 = 750 m. A solved distance off by exactly 1000× is the signature of this slip.
Enter the dimensions so they store
Three habits prevent a repeat and protect every later step that consumes the numbers.
- Measure the printed board, not the drawing. Printers rescale, so a nominal 50 mm marker can print at 49.3 mm. Use calipers, then convert. The full procedure is in Measured marker size in hand-eye calibration: print and measure.
-
Type dots whatever your keyboard layout suggests. The parser wants
0.0502; give it that. -
Read the echo back. A healthy run reports
marker_size 0.050200andmarker_seperation 0.071000with dots. Zeros or commas in that line mean the values did not survive.
The same locale category bites later in the pipeline. Numbers written to a calibration file follow LC_NUMERIC too, and a file of comma decimals fails to load elsewhere — Camera calibration file error? Check locale number parsing covers that failure. Panels that quietly revert saved settings share the root cause: see Camera settings saved but not applied at open: fix the locale. Numeric comma bugs are not unique to vision tools either; HMI and control software fails the same way, and the Ignition decimal point and comma locale fix walks through the identical mechanism on the automation side.
When it is not the locale
Some adjacent failures look similar and need different fixes.
- Freeze on topic selection with genuinely empty fields — if you never typed values, or the fields reset on a dot-decimal machine, the crash path is different: Calibration toolbox crashes on topic selection: fix zero marker size covers that one.
- Board detected but the pose never converges — a detection or solve problem, not a number-format one; start with ArUco marker hand-eye calibration: why the marker pose stays unknown.
- Solve finishes but the hand lands wrong — check the transform chain next; Hand-eye calibration orientation error: fix the quaternion mismatch is the usual culprit.
A clean run looks like this: CameraInfo selected first, then the image topic; the terminal echoes marker_size 0.050200, marker_seperation 0.071000 with dots; the board highlights in the image; the panel stays responsive.
Sixty-second recovery
- Kill the frozen RViz process; the panel does not recover in place.
- In the launch shell, run
localeand readLC_NUMERIC. - If it is not a dot-decimal setting, run
export LC_NUMERIC=en_US.UTF-8and confirm withlocale. - Relaunch RViz from that same shell.
- Select the CameraInfo topic, then the colour image topic.
- Type
marker_sizeandmarker_seperationin metres with dots:0.0502and0.0710. - Re-select the image topic and check the echo: dots and non-zero digits mean the parse survived.
From there the calibration itself takes over: capture poses, solve, verify against a known distance. If you are speccing hardware for a new cell instead, the camera finder filters GigE and USB3 models by resolution, the system configurator builds the chain around them, and more fixes like this sit on the cameras topic page. For a question this page does not answer, the contact page reaches the applications team.
Questions
Why does my solved marker distance come out 1000 times too far?
Almost always a unit slip: millimetre readings entered as whole numbers, or a comma that swallowed the fraction. A 50.2 mm marker entered as 50.2 is 1000 times too large, and every solved pose scales by the same factor, so a marker truly 0.75 m away reports 750 m. Check one known distance after solving; an error of exactly 1000x means retype the entries in metres with dots, not a recalibration.
Do I have to change my whole system locale to en_US?
No. `export LC_NUMERIC=en_US.UTF-8` changes the numeric-formatting category only, for programs started from that shell. Desktop language, dates and messages stay untouched, and the setting disappears when you close the terminal. If you calibrate often, keep the export in a short launcher script so every session starts with a dot decimal without touching system settings.
Should the CameraInfo topic be selected before the image topic?
Yes, it is the safe order: an empty camera info can produce its own all-zero behaviour. It will not fix a parse failure, though — if the dimensions still print as 0,000000 with commas after CameraInfo is selected, the typed values never converted, and the fix is LC_NUMERIC, not the topic order.
What exactly do I measure for marker_size and marker_seperation?
marker_size is the side length of one printed black square. marker_seperation is the distance between the origins of two adjacent markers, centre to centre, not the white gap between them. Measure the physical board with calipers after printing, divide millimetres by 1000 for metres, and type both values with a dot.
Can the same locale problem corrupt the saved calibration file?
Yes. Numbers written to disk follow LC_NUMERIC too, so a capture made under a comma locale can save comma decimals that other tools refuse to load. Run the whole pipeline — capture and save — under LC_NUMERIC=en_US.UTF-8, then open the file and check that a value such as 0.050200 appears with a dot and six decimals.