Bildverknüpfungen... Wie viele Bilder?

Question

In Image Links steht, dass "die Steuerung mit mindestens 2 Bildern verknüpft sein muss, aber 3-8 Bilder am besten sind". Was tun Sie also, wenn Sie 15-20 verfügbare Bilder haben? Sollten Sie Verknüpfungen in allen herstellen oder bis zu 8 auswählen, und wenn ja, wie wählen Sie diese 8 aus? Zum Beispiel die ersten 8 auswählen, diejenigen mit der besten Qualität, versuchen, eine gleiche Anzahl zwischen verschiedenen Winkeln zu wählen usw.?<\/P>

IaGeoff · Answer

About "Which" ones to select.... if you're higher altitude (thus, more images available), then focus on Alternating (every other one) along a flight line.... and also definitely include Parallel Flight lines (side-to-side) so you have a true Stereo.  Also, pay attention to your flights Overlap settings and use that visually where a GCP and Check Point is situated within Image.

You mentioned "attempt to choose equal number between various angles", so assume above would cover.  Also note your Flight line transitions (Turns/Spins/Etc.).  Sometimes these are good to include, sometimes not... depending on the Images orientation with others... that's why you'll see some of the Images in the AUTO LINK step just don't even come close.  So why an important step to Validate each Image's Marked Link position with Target/Feature/Etc. and also look at any Image factors to weed out.  Best to classify as a Check Point (CP) vs. a GCP that Constrains Models..... and THAT's really the key point.  You'll learn quickly what makes a good model using distribution of GCP's...... yet if you want to support the work, make sure you have ample Check Points distributed as well to support the Processing Results.

Suggest always strive for 3 or more Images per Control feature (while 2 is min., just doesn't fully equate to the math involved), and wouldn't exceed 8 to10 (12 max).  The time your PC takes to "Match Feature Points" will be a sign on just how "Constrained" you've taken Images to be included for Primary Control (GCP's).  Overall, focus being.... what are you trying to Model ?  This can come MORE into play if you want to Model say a Structure and Capture more sides and Perspectives (3D Model aspect) around an object.  Normal Aerial Stereo Matching (Camera straight down), the 6-8 Images is ample.

Lower Altitude, less Control displayed in Images yet product Details/Results generally increase and you'll end up with more Matching details.

Hopefully helps.

IaGeoff · Answer

Chris,

Great ESRI resource is https://doc.arcgis.com/en/drone2map/latest/help/control.htm#GUID-A1765DA9-EBD4-4765-BE45-7A63B60DE480

Simply go down to the "Image Links Editor" Section of page, and basically outlines.

Not sure if you were able to outline, yet hopefully this may help if not:

a.) In the Image Links Editor, what they refer to is Match Images to GCP's and Check Point's for "EACH" Control Point... so the statement about # of recommended Images Matching "Each" specific Control Point as you mention.  (ESRI should use CKP/CHP or something ELSE, not CP which is an Industry Standard for Control Point..... yet the Report does state as "Check Point" !).

b.) Instead of going through each Image and Finding your Target/GCP (or, user could use a Photo Identifiable Known Point) and digitizing.... simply just pick "ONE" Initial Image with your desired Control Point and Digitize (Select Center of Target/Feature in Image/Spot/etc.).    Then, select ESTIMATE LINKS.... this will pull up all Images with that spatial location.  Filter by "Distance" on top to have images listed in order of Distance from your Initial location selected.  Then, select AUTO LINK which will place a Mark on each of the Images from the ESTIMATE LINKS step.

c.) NOW... the key is user STILL needs to go through the List of Images for EACH Control Point and Verify two things:

1.)  You want to use that Image for Control Link

2.) that the MARKED point generated during AUTO LINK is correct, which for a number of Images it will NOT... with some Off, Way Off, or WTF.  Update the Marked Link simply by zooming in and Digitizing the correct location for each desired image.  You'll see a YELLOW mark placed at location, so ensure when you select new Point.... the Yellow Mark is displayed !!  And remove those Images that just don't support the Image Link (Match Point).... maybe too close to Image Edge, in a Spin/Turn for sUAS capture, etc.

So, basically that's it.... Each Control Point (GCP and Check Point) you define in Project's Control needs to be conducted, so that you get the IMAGES to Constrain to the GCP's and report you the RMSE.... and Check Points will report RMSE results as well, being a CHECK from the Models results at that location... all within the Processing Report under "Geolocation Details".

ALSO, take the Step to EXPORT your CONTROL after conducting the Image Linking step.  Save you a TON of time if you want to use same Images and Control in another Project /Recreate/etc. !!  Very important legacy archive step NOT to miss.

Use for every project with a combination of Static Survey Primary CP(s), Propeller AeroTargets, then just nRTK (Network RTK GPS) Obs. on Photo Identifiable (or Painted Targets) that I choose to use as more GCP's or just Check Points.  Results are always tight.... ultra tight !!

Hopefully above helps outline.