Different Camera Height - Drone Height

[Mapcon_RL]

Hello Everyone,

Just logged in, so what i ask now might have been reported already but i've found no proper answer to the subject.
Considering the post http://www.agisoft.com/forum/index.php?topic=2876.msg15172#msg15172 i'm also trying to align images taken from different altitudes.
As you can see on the page print, i had to fly different plan with multiple heights. Now, on the processing move, i keep having
these misaligned (Z) for the area!
Anyone could help?

Thanks

RL

[SAV]

Hi Mapcon_RL,

If you were flying different plans at multiple heights, then I cannot see any problem in the screenshots you shared. That's exactly what it shows.

If the overlap between the different surveys/flights is sufficient, then you should be able to process everything in a single chunk.

In case I misunderstood your post and you are are having issues with the altitude values (Z) derived from EXIF data, then have a look at this post:
http://www.agisoft.com/forum/index.php?topic=7309.msg35184#msg35184

Regards,
SAV

[Mapcon_RL]

Hello SAV, thanks for the reply!

The point is to get an equivalent/equal surface height - in order to build a DTM, from flights that have been taken from different altitudes.
ie. "If the overlap between the different surveys/flights is sufficient, then you should be able to process everything in a single chunk." - Yes, it is enough, but doing it on the "normal" way i got those differences from the last pics print also over the dense cloud product...Is there a procedure i could do either than the ones pointed on the post http://www.agisoft.com/forum/index.php?topic=7309.msg35184#msg35184 ?
Considering the post http://www.agisoft.com/forum/index.php?topic=7309.msg35184#msg35184, i understand what you meant. I chose the first and second option from there, selected 200 meters accuracy and unchecked for cameras references. Now seems that the cloud better represents the surface even though the pictures are not evenly aligned.
I have GCP which could help me to fit that on the real world... Is that right?
Do you think this 200 meter accuracy would cause me any hassle onto the process that would be better to avoid?
I first thought about changing the heights from EXIF...however, by doing that i thought it could generate errors of measurement and scale...what a bout it?

Thanks

Best Regards

RL

[SAV]

Hi RL,

If you have GCPs, then make use of them. 

It is probably a good idea to fix the altitude value in the EXIF information, because then it can also be used as a reference. In your case, a simple Python script I shared in an earlier post should work quite well.
http://www.agisoft.com/forum/index.php?topic=4986.msg40010#msg40010

You just need to know the elevation (above mean sea level) of the point of take-off for EACH individual survey plus the flight height of each survey (the value you entered in your mission planning app). You could use this website to figure out the elevation value from where you took-off: http://elevationmap.net

Follow this workflow
1. Create a chunk for each flight (= 3 chunks in your case)
2. Run the Python script for each chunk to adjust the exif Z value
3. Move the images/cameras from chunk 2 and chunk 3 to chunk 1 by selecting all cameras/images in chunk 2, then right click and choose MOVE CAMERAS > CHUNKS > Chunk 1. Repeat for chunk 3. Now you should have all height adjusted images/cameras in chunk 1.
4. Run alignment
5. Import and 'link' GCPs/markers (make sure that the coordinate reference systems of markers and images are matching). Double check the accuracy settings for cameras (10m default) and markers. If available, also import the horizontal and vertical accuracy values for your markers. If they have been 'properly' surveyed, these should be available.
6. Remove 'outliers' from the sparse cloud (i.e., using gradual selection) and then run OPTIMISATION
7. Build dense point cloud and other spatial datasets (mesh, DEM, etc)

DONE.

All the best.

Regards,
SAV

[Mapcon_RL]

Hello SAV,

Thanks a lot for the workflow.
It all have helped a lot. I followed it as you said and it aligned the height for the whole lot.
After that, i followed the other workflow you posted http://www.agisoft.com/forum/index.php?topic=7851.msg37494#msg37494...
That all came very easily, however, at the end i got few pics misaligned, as the print attached.
I tried to realign it but none of these have succeed. Is there a tip over this alignment feature we could you to ease the way PS aligns pictures?
All of those were taken with 70% overlap - side and front, and still some won't align!
Thanks a lot.

RL

[SAV]

Hi Mapcon_RL,

Your scene looks quite complex, hence I would suggest to use a higher forward and sideward overlap next time. Are the pictures that cannot be aligned much closer to the object, for example are they over an area with large trees whereas the rest is over an area that does not include high vegetation? This would mean that the image overlap in these areas is actually much lower, hence the image alignment is struggling. The front and side overlap that you set up in your mission planning app assumes that you are mapping on a flat surface at an elevation that is the same as your point of take off. Only along this level, the overlap is what you have set it to. If the object is higher than the ground (e.g., tree, building, ...), the overlap is smaller. If the object is lower as the ground (i.e., a hole), then the image overlap is higher. Keep that in mind when planning your missions.

To fix your alignment issues you could try to insert markers on image pairs to assist PhotoScan. Generate a few markers on a aligned image (you don't need to enter any coordinates for it) and then mark the same features in the not aligned image(s) using the same marker ID. This will generate 'manual tie points' that should help PhotoScan to figure out the camera position and orientation.

All the best.

Regards,
SAV

[Mapcon_RL]

Hello SAV, thanks for the tips.
For sure what you said is true. The complexity of the scenes, despite the high overlap. Also the area has to many different size trees, which sometimes brought the camera too close to the target.
Some other areas within this project we flew higher and had no complications...
Regarding the manual alignment, after marking the points i shall only optimize the cameras and hope to get it done, right?
Once again, thanks a lot.

RL

[SAV]

Hi Mapcon_RL,

No worries. I had the feeling that high trees might be the reason for problems. Slight movement of branches/twigs will lead to alignment issues because the object has changes from one picture to the next one and cannot be accurately matched.

Also, sometimes if the quality of the imagery is not great (i.e, blurry imagery), you might get a better alignment when running it at MEDIUM quality where it only uses 'downscaled" images. Test if quality MEDIUM works better than HIGH. But generally I suggest to use HIGH. Note that you can check the image quality by right clicking on one of the pictures and then choose ESTIMATE IMAGE QUALITY. You should not use any pictures below 0.5. Everything above 0.7-0.75 is generally OK.

One more thing. You could try to increase the key point limit to 80,000 (instead of default 40,000) to improve feature detection in the images that is needed for image alignment. Processing time will linearly increase though.

You'll have to re-run Photo Alignment after inserting your manual markers and then optimise again.

All the best. I hope you/we can get your project successfully reconstructed. 

Regards,
SAV

[Mapcon_RL]

Hello SAV,

Thanks again for the advice.
I was just about to inquire regarding the possibility of "downsampling" images in order to get better alignments...
I've tried all type - Lowest till highest and they return basically the same amount os misaligned pictures.
Ive also estimated quality for them and they all over 0.82, which seems OK. Despite increasing the point limit to 80k, is there any other tip such changing the parameters for the accuracy in mts and dgs?
I've tried to align these same images with DroneDeploy and Pix4D and they all have done it. Into DroneDeploy i couldn't control the parameters so i cannot say much. However, in Pix4D, i've changed the scale for the pictures to 1/2 and set to an "Alternative mode" for the alignments...Is there any thing comparable onto Agisoft 1.4?
I'm sorry for the comparison and the whole bla bla bla, but i'm testing it all and trying to make a point out of it.
BTW, concerning the results i've had so far, the job done by Agisoft in order to generate DTM, clean up the vegetation and so on is awesome. Clap hands to the results, which are better in accuracy and positioning than these other two softs i've tested.

Thanks Again

Best Regards

RL

[SAV]

Hi Mapcon_RL,

The 'downsampling' of images is done by changing the quality settings when running the Photo Alignment in PhotoScan.

ULTRA HIGH = uses photos at 200%
HIGH = original size of photos (100%)
MEDIUM = photos at 50%
LOW = photos at 25%
LOWEST = photos at 12.5%

As far as I understand/know, choosing scale 1/2 in Pix4D is basically the same as choosing quality MEDIUM in PhotoScan.

Image quality of 0.82 and above is very good, so that's definitely not the issue.

Have you tried to place manual markers for the pictures that weren't aligned? In such pictures, you would not use a feature on a tree but something on the ground that you are sure of is stationary/not moving. That has worked for me in the past quite well.

One last resort. Try to also increase the tie point limit (or leave empty to use the maximum number of tie points).

The advantage of Agisoft PhotoScan is that it gives you much more control over processing parameters compared to other software packages or cloud computing platforms such as DroneDeploy etc.

All the best.

Regards,
SAV

[Mapcon_RL]

Hi SAV,

OK for the downsampling figures, i actually thought would be something like it but now we know the parameters.
I placed manual points but it didn't help a lot. The "optimize cameras" has been thru and didn't align them. After that, i run the "Reset and Align Photos", but yet no luck. Even tried to realign all the pics on the chunk maintaining the manual markers...no luck either.
I read somewhere that keeping 0's on the point limits would allow PS to run as much points as possible and also done that....no luck.
Would have a way to import into PS these aligned photos from Pix4D?
Indeed, to me this is a + + advantage of PS. Don't like the idea of not knowing or control much of the tasks performed by software. This is the main reason i'd still stick with PS rather than any other.
BTW, would you know which height parameters are used for those different classes while classifying the dense cloud? What are the heights associated to High Vegetation, Medium Vegetation and so on?

Thanks again

Best of all