What is 3D Camera Calibration?
3D Cameras are optical instruments. They can measure the depth of each pixel through a variety of methods (Stereo, Structured Light, LiDAR, ...). Each of these methods is founded on basic physical modeling of the sensing situation. For example if we want to measure depth through Stereo, we have to know exactly how far our cameras are spaced apart and what their relative orientation is. All the knowledge of this internal camera state we call the "internal model".
However, sometimes the physical reality slightly differs from our internal model, which leads to systematic measurement errors.
3D Camera Calibration is the process of changing the internal model so that it aligns again with the physical reality, with the goal of minimizing systematic measurement errors.
The most common systematic measurement errors are:
- Distortions in the depth map. I.e. the 3D image of a flat wall is not completely flat. The flat wall appears slightly curved or wavy.
- Scale errors: All things are measured too close or too far away.
- Field-of-View errors. The imaged scene is slightly warped. Angles between surfaces are different depending on the camera orientation angle. Measurements in the XY plane are off.
- Tilting: The image of a flat wall that is faced orthogonally appears tilted / at an angle.
How does DotProduct perform 3D Camera Calibration?
DotProduct has developed proprietary calibration technology to minimize systematic measurement error to a degree that goes beyond what the factory calibration and/or manufacturer tools provide.
DotProduct's calibration procedure depends on the camera model used.
- Intel RealSense D400 series:
- The camera is checked for existing measurement errors in various conditions
- RealSense
"Dynamic Calibration" is carried out if necessary to take care of gross
errors. New internal model is stored on-camera.
- DotProduct's
calibration tools are used to refine the internal model and to address any remaining systematic measurement errors. Results are
stored in a calibration file that can be used within Dot3D.
- Primesense / DPI-10 / DPI-8
- DotProduct's
calibration tools are used to determine the internal model and to address any remaining systematic measurement errors. Results are
stored in a calibration file that can be used within Dot3D.
How does DotProduct's calibration differ from other calibration tools?
DotProduct's calibration methods are specific to each camera model to achieve a maximum of accuracy. Beyond the typical calibration parameters (intrinsics, extrinsics), DotProduct's calibration can recover additional parameters of a camera's internal state that can be used to minimize systematic measurement errors.
Also, in contrast to regular / factory calibration the result of DotProduct's calibration cannot be stored on-camera. Instead the results are stored in a file that can be loaded within the Dot3D software.
Depending on the distance to the target, using DotProduct's internal calibration we are able to achieve typical Z errors of 0.2% - 0.8% and XY errors of 0.1% to 1.2%, vs. typical Z errors of 1% to 2.5% and XY errors of 0.5% to 3% when using only factory calibration (numbers refer to measurements from a single camera viewpoint and for ranges up to 3.5 meters).
In addition DotProduct's internal calibration significantly reduces local surface bending / Z-distortion.
How to perform 3D Camera Calibration for Intel RealSense D400-series cameras?
Intel RealSense offers two calibration methods for D400-series cameras to provide a baseline of accuracy and to get damaged or otherwise out-of-calibration cameras back into shape.
The two methods are "Self-calibration" and "Dynamic Calibration". The Self-Calibration method itself has several components. For most cases DotProduct recommends the Dynamic Calibration method.
Here is a high-level overview:
- Self-Calibration (available in RealSense Viewer)
- "On-Chip Calibration"
- Intrinsics
- Extrinsics
- "Tare Calibration"
- Dynamic Calibration (separate app, using a physical / printed target)
The Self-Calibration is available through the RealSense Viewer. More information about Self-Calibration is available here:
The Dynamic Calibration is a separate app. The latest version, as well as more information can be found here:
Even though the Dynamic Calibration takes a bit more time to carry out and requires a printed target, we found that it is quite reliable and relatively easy to do. We are likely going to post a video tutorial about this in the future.
The Self-Calibration on the other hand is extremely easy to perform. However, for certain measurement / depth errors and/or cameras that are more than slightly out of shape we found that it is not able to calibrate as reliably as the Dynamic Calibration.
We therefore recommend the "Dynamic Calibration" method.
For a maximum of accuracy we always recommend a calibration by DotProduct using our internal tools as described above.