Best Practices and Effects of Non-Ideal Chart Framing

A comprehensive study of the accuracy of distortion and field of view measurements in Imatest from a variety of test targets/analysis under varying degrees of barrel/wave distortion, tilt, rotation and translation has yet to be performed and is in our medium-range backlog.
Once this has been completed, it will allow us to give more specific suggestions based on users individual setups.

Until then, the best practices for FoV ≲ 140 degrees are as follows:

Use a high frequency checkerboard
1. More points to measure → more accurate fit
2. Limit: Squares must remain large enough that accurate detection near the edges of the image is possible
Align the checkerboard as perfectly as possible
1. Chart must overfill camera FoV
2. Position chart perpendicular to optical axis of camera
3. Get the “Edge angle” for squares as close to 0deg as possible
  1. For highly distorted images, focus on edges in the middle row and column. Edges away from the center will have some angle due to the distortion.
4. Chart should be centered in the image
5. Adjust the camera-to-chart distance to position detectable saddle points as close to the edges of the image as possible
6. Here is an example of the ideal framing for an image with minimal distortion.
  Note that you would want something with higher frequency than this.
Uniform illumination
1. No saturation in the middle
2. Bright enough at the edges to have high enough contrast for detection

For FoV between ~140-160 degrees, or FoV large enough that it is difficult to overfill the FoV with the checkerboard chart, we believe there may be some benefit to using a pre-distorted SFRplus target instead, but this is something that needs to be tested further when we do the testing mentioned above.

For FoV > ~160 degrees, our calculation starts to break down and we would not expect accurate results for images with this much distortion. We expect to be able to give a more specific value for where this breakdown occurs once we have completed the testing mentioned above.

For now, the best information we can provide on the influence of non-perfect alignment of the chart is the following -

These results are from measurements of small batches of images from a single camera where I was trying to get a rough estimate of the effects of chart rotation on the FoV measurements in Imatest. Note that this was a quick test, so I did not do a good job of isolating chart rotation as the only dependent variable, and I did not have a large enough chart to fully fill the FoV, so the diagonal FoV requires significant extrapolation in each case.
Here’s an example of a few of the images:

For the measurement, I took batches of 15 images at 6 different rotation angles, analyzed them in Imatest to get the vertical, horizontal, and diagonal FoV results, than ran statistics on the results for each batch of images to obtain the plots shown below.
In the following figures, the x-axis in each case is the average edge-angle calculated for each batch of images, and the y-axis is the FoV shown as a box-and-whisker plot which is explained here:

Note that, for these images, the FoV measurements are mostly consistent between -5 and 5 degrees of chart rotation. This is why I recommend getting the average edge angle as close to 0deg as possible for your measurements.