Q & A | September 2008

Automatic interior orientation (AIO)

What is AIO?

AIO is an automated step in the photogrammetric workflow designed to increase productivity and accuracy for interior orientation of aerial images. SOCET SET employs AIO to speed manual interior orientation (IO). The underlying algorithm performs rigorous batch processing on multiple images being scanned simultaneously, eliminating the need for manual measurements.

The IO process for Frame and Panoramic sensor models converts film space (camera focal-plane space) to digitized image space. The units in film space are millimeters, whereas the units in digitized image space are pixels. Therefore, the transformation from film to image space is critical. The location of all fiducials is used to derive the transformation parameters for successive applications, such as triangulation or DEM generation. Inaccurate IO results decrease the final accuracy and reliability of related applications.

AIO is widely used in SOCET SET today, and will be available to SOCET GXP users in the near future. This article — although a rare case — examines an interesting scenario that may help other users understand the AIO mechanism.

Case-study example

While investigating an unusual AIO problem for a customer — as indicated by the AIO log files, roughly 25 percent of the frames failed with AIO operation among approximately 1,600 images — the GXP team recognized that the sup files were contaminated, thus advised the customer to re-import the images. However, the re-import yielded similar results. On additional investigation, the GXP team found that lowering one value (FOM_LIMIT) of one strategy file can help correct AIO results.

Interestingly, although this approach yielded good results, there were additional anomalies. The debugging results indicated that the scan resolution derived from the sup files, before and after AIO operation, was an approximately 0.7 micron-per-pixel difference. Before AIO located the fiducials, the scan resolution was 13.3 microns per pixel, while the scan resolution changed to 14 microns per pixel afterward. This had never happened before, and of course, should not happen at all.

It is well known that the default frame size is 230 mm x 230 mm for all commercial aerial cameras. Presuming that the scan resolution is 14 microns per pixel, the digital frame size is approximately 16,428 x 16,428; 0.7 micron per pixel difference. A scan resolution of 13.3 microns per pixel will result in a different digital frame size: 17,293 x 17,923. This indicates that the digital size is a difference of 865 pixels (17,293-16,428 = 865 pixels).

The customer’s scanner provided each frame with about 17,293 x 17,293 of the frame dimension for one of the commercial cameras called RMK TOP 15. The first scan resolution is based on SOCET SET’s frame-import, 230,000/17,293 = 13.3 micros per pixel, which helps initiate AIO’s automatic processing. The second scan resolution of 14 microns per pixel derived after AIO correctly located fiducials, should be much more precise than the first one.

However, considering the second scan resolution and the digital frame dimension of 17293 x 17293, the frame size was approximately 242 mm x 242 mm, quite different from the default of 230 x 230, which indicates that the sup files are contaminated — as initially suspected. Thus, it appears that AIO processing failed for some of the images.

Q&A

How did this scenario happen?

The scan range was set incorrectly. Always check the scanner to make sure the scan range is set correctly and that it matches the frame. Note the difference between the yellow and red markings in Figure 1 below; red is wrong, yellow is correct.

Figure 1: Make sure the scan range is set correctly before scanning; red is wrong, yellow is correct.

Figure 1: Make sure the scan range is set correctly before scanning; red is wrong, yellow is correct.

What kinds of problems does this scenario cause?

The AIO operation needs more time or simply fails since the scan resolution and camera type are only a priori knowledge that the AIO relies on to implement fully automatic processing; the wrong scan resolution means that the AIO starts searching blindly from an incorrect spot, and thus needs to work harder. Alternatively, the operation may have failed due to bad approximations. See Figure 1.

Is it possible to recognize the problem?

Yes. Since the scan resolution is normally set by the operator, perform a simple math calculation to guarantee that the scan range is set correctly; 230,000/scan resolution for the initial frames scanned, to verify that the scanned images’ dimensions are close to this value.

How do you resolve the problem?

It is not necessary to re-scan. Simply change the default size from 230 to an approximate value, such as 242 during frame import. See Figure 2.

As noted for the example cited in this review, no coding change is required for SOCET SET v5.3.1 or later versions. After changing the default size from 230 to 242 during frame import, the AIO ran very quickly and smoothly with a 100 percent success rate. See Figure 3.

Figure 2: Make sure the scan range is correctly set before scanning; red is wrong, yellow is correct.

Figure 2: Make sure the scan range is correctly set before scanning; red is wrong, yellow is correct.


Figure 3: In case the scan range is set incorrectly, calculate the real scan range based on the scan resolution, and then change the size(X)/size(Y) during frame import to guarantee a correct IOCEFF matrix in sup files for AIO operation.

Figure 3: In case the scan range is set incorrectly, calculate the real scan range based on the scan resolution, and then change the size(X)/size(Y) during frame import to guarantee a correct IOCEFF matrix in sup files for AIO operation.

Comments are closed.