• Test F3F3 - Analysis

Test F3F3: Clark and Tikir Photometric Function - Analysis

Since the Clark and Takir photometric function behaved so differently to the Lommel-Seeliger photometric function during testing, the sub-test results are analyzed here in detail. The F3F3 (Clark and Takir photometric function) digital terrain model suffered from pervasive degradation with every processing step performed, eventually failing to align/correlate images for a subset of landmarks at the 5cm tiling step. The minimum image correlations for each 5cm landmark are plotted for the F1F1 (Lommel-Seeliger photometric function) and F1F3 (Clark-Takir photometric function) tests. The failure mechanism is illustrated by the 5cm Landmark Test, reported below.

5cm Landmark Test

Background: A subset of new landmarks (28 of 225) created during the F3F3 subtest 5cm tile failed to achieve minimum image correlations across the images. At 5cm each landmark contains approximately 12 Detailed Survey PolyCam images. At this scale elimination of low correlating images is not allowed, instead low-correlating images must be aligned by hand. On investigation it was found that the lithos template degraded rapidly with iteration leading to rapid loss of image correlation scores. This behavior was found to be pervasive across landmarks of all latitudes within the TAG site region, all available viewing conditions and all ground sample distances (GSDs) including the start model landmarks at 35cm, 75cm, and 150cm GSD. The behavior is illustrated from the performance of this additional test.

Methodology:

• Working directories were created for the full suite of photometric function tests (F3F1, F3F2, and F3F3).

• The working directories were synced with the 10cm pre-geometry back-up for each sub-test.
• The geometry processing step was performed for each subtest and each subtest model achieved the Residualed Processing Step.
• A 5cm tiling bigmap was generated as per standard F3F processing. Showmap was used to generate an image of the tiling bigmap.
• Using LITHOS (with the appropriate photometric function) a 5cm landmark was created at the center of Tag Site #1 (centered on -8.027, 262.768, lat, wlong, q-size of 49).
• The standard 5cm tiling lithos commands were followed to add images, turn on extract filter, get heights from tiling bigmap, find normal from heights, find Z from V and pointing, and auto-eliminate images.
• An auto-image-alignment was performed and the start correlation scores recorded.
• Using all images, a template build was conducted using 50 iterations and the output landmark display was recorded.
• The last two steps were performed iteratively until 1000 template building iterations had been performed.

Results and Discussion:

The tiling bigmaps, lithos image correlation scores and lithos output landmark displays are presented for subtests F3F1 (Lommel-Seeliger photometric function) and F3F3 (Clark and Takir photometric function).

Poor performance of digital terrain modeling prior to the 5cm tiling step is clear from a visual comparison of the tiling bigmaps which contain 10cm, 20cm, 35cm, 75cm, and 150cm data.

Comparison of the lithos image correlation scores through processing shows that the F3F3 subtest starts with moderate image correlations which degrade to low/no correlation with template building, whereas the F1F1 subtest starts with high correlations which improve to excellent correlation throughout template building.

A visual inspection of the output landmark displays show how quickly the F3F3 template degrades, and how poor it becomes. Interestingly, a subset of images in the F3F3 landmark maintain correlations scores between 0.4 and 0.5, despite no topography being apparent in the template.

Images of the 5cm Tiling Bigmaps (F3F1 left, F3F3 right):

LITHOS Image Correlation Score During Template Building:
F3F1 - Lommel-Seeliger Photometric Function:

F3F3 - Clark and Takir Photometric Function:

Start (F3F1 left, F3F3 right):

Template Building: 50 Iterations (F3F1 left, F3F3 right):

Template Building: 100 Iterations (F3F1 left, F3F3 right):

Template Building: 150 Iterations (F3F1 left, F3F3 right):

Template Building: 200 Iterations (F3F1 left, F3F3 right):

Template Building: 250 Iterations (F3F1 left, F3F3 right):

Template Building: 300 Iterations (F3F1 left, F3F3 right):

Template Building: 350 Iterations (F3F1 left, F3F3 right):

Template Building: 400 Iterations (F3F1 left, F3F3 right):

Template Building: 450 Iterations (F3F1 left, F3F3 right):

Template Building: 500 Iterations (F3F1 left, F3F3 right):

Template Building: 600 Iterations (F3F1 left, F3F3 right):

Template Building: 700 Iterations (F3F1 left, F3F3 right):

Template Building: 800 Iterations (F3F1 left, F3F3 right):

Template Building: 900 Iterations (F3F1 left, F3F3 right):

Template Building: 1000 Iterations (F3F1 left, F3F3 right):