How Can We Test the Accuracy of A 3D Scanner?

When a 3D scanner user wants to test his scanner, usually he will concentrate on two indicators:

  1. Accuracy - Whether the dimension is accurate.

  2. Details - Whether the details meet the expectations.

Today, we will focus mainly on how to measure the Accuracy.

Typically, professionals in the 3D scanning arena use standard measurement tools such as a Ceramic Standard Ball, Ceramic Calibration Plate, Step Calibration Block, etc.

Because these measurement tools have specific known attributes and dimensions, accurate results can be obtained.

We will now describe the use of Ceramic Standard Ball to test accuracy.

This is one of the standard measurement methods.

Some users may ask ”I don’t have the measurement tools, but I do want to know my 3D scanner’s accuracy level. What can I do?”

Some use a caliper to measure the subject and compare with the dimensions of scanned model files.

If you don’t have the standard measurement tools (or the true values are unknown), that method can suffice.

However, we have two suggestions:

  1. When you measure the dimensions of an object with a caliper, you should use the average values after multiple measurements to avoid errors due to manual operation of the tool.

  2. When performing dimensional analysis on the scanned model data, it’s better to choose Fit Cylinder or Fit Plane, rather than randomly selecting two points of the mesh as your reference.

Here are some examples:
Check these pictures and guess which are accurate?

Yes, the first and second one use the correct method, while the third one is questionable.


Because when you randomly choose two points, you cannot ensure that the two selected points represent the true distance between those two features.

In conclusion:

  1. When you want to test the accuracy of your 3D scanner, we recommend you to use standard measurement tools.
  2. If you don’t have these tools, then you can use a caliper, but remember to take multiple measurements and average them.
  3. When analyzing the model data, do NOT randomly grab two points of the mesh directly, as the result will probably be inaccurate.