Different lighting in images

I am having trouble with a survey that was flown over a couple of different days, most of the survey was flown on quite a bright day but some was flown on a more overcast evening, the images have noticably darker exposure but the images are still clear and good. When the survey is processed howevwr there is a definate step line along the different exposure images, probably a step of over 1 meter. Is there any way to work around this issue?
Thanks

Good question Lee, I had exact same problem, I send message to support, their lovely answer was that I did not have sufficient overlap between flight blocks, which did not makes sense, as I lots of overlap. They also said I should add more GCPs or MTPs, too many GCPs was just not cost effective for us, MTP were not so easy to use as its wilderness, so hard to mark objects. Iā€™m not convinced that it is software issues. What I tried to do then is, to edit the images using ā€œimbatchā€ free software,which allows you to batch process them. So, a tweaked Contrast, Brightness and temperature. The I went ahead and process them in Pix4D. This helped in some flight block but few of them were reflown as I could not fix them. Lets hear what other say about this.

You might try taking your images into something like Adobe Lightroom. Lightroom lets you view a library of images all at once as on a light table and make adjustments to individual photos, groups of photos, or all of them. Changes applied to one photo can be applied to others.

I might select two representative photos from each of the different sets and see what I needed to do to balance their differences while preserving as much information as I could. You donā€™t want to clip either the darks or the lights when making those adjustments, else lose information in those areas.

You mention a ā€˜step lineā€™ between image sets, but itā€™s not clear to me what exactly that step is. Horizontal displacement? Vertical? A bad looking blend line in a mosaic?

If a technician has a difficult time locating a tie point so might the software. How much overlap you really need depends on the subject but a lot on your lens as well. The super wide angle lenses can produce tons of overlap but a completely dissimilar perspective that doesnā€™t match easily.

Some combination of additional tie points along the seam line between image sets could help, and or processing them as separate areas with those common tie points. Masking of one line in the overlap area could help too. Depends on whatā€™s wrong. But my first stop would be working on the imagery.

Steve

Hi Steve

You are making a good point here. I normally use Voitglander Helliar II or III, 15mm lens. I like it because it is wide as you have pointed out but most importantly I like the true infinity focus. Anyway, toĀ  your question, all my displacements were vertical one block was off by over 7m from the rest, in the DSM you could seen a cliff which in real life does not exist, not sure Lee saw perhaps similar thing. Lightroom is a good point too, I have used it to try to correct vignetting, not all that successful, but it helped to some extent. But I decided to increase side overlap which seems to fix the vignetting problem.

Ā 

Clarence

Hi Steve, Clarence
Yes I have a verticle offset as well, maybe about 1 meter, overlap is not an issue as you can see overlapping data dissappearing below, its like an overlap in point cloud data with one higher that the other, namely the darker imagery. I might try what you suggested with lightroom and see if that works, its dissappointing as pix4d say exposure levels shouldnt effect results.

Steve,
What do you mean by masking off an area?

Clarence, 7 meters? If you are holding onto good ground control I donā€™t see how that is possible. Althoughā€¦

Have you taken a look at your camera model? Iā€™ve seen a lens get assigned parameters that werenā€™t correct based on the softwareā€™s reading of theĀ photoā€™s metadata when using manual or aftermarket lenses on camerasā€¦

Pix4D did what it could with the exif data. IĀ think the default was a 35mm focal length though Iā€™ll have to check. The big trouble came with a default to the fisheye camera model. It didnā€™t occur to me to look for that the first time and took a while to figure out. But you can definitely get some funky vertical values if the focal length is off. And just generally bad results if other parameters are no good. Worth checking

Optics have come a huge distance in the past couple years. That Voitglander looks like a sweetie. Iā€™ve got a wide angle Rokinon prime lens on my Nikon.Ā Similar field of view. Manual focus as well. When dialed in the thing is breathtaking. I shoot everything manual with focus fixed and on one f-stop where depth of field, aberrations and sharpness are allĀ in their happy place.

Steve

Ā 

Ā 

Lee,

Iā€™m not sure whereĀ youā€™re quoting from Pix4D about exposure levels not affecting results as its all about good exposure. Iā€™m thinking whatever was said is meant to be interpreted as a consistent exposure across images that could be relatively lighter or darker to a point, provided there wasnā€™t gross under or overexposure with data loss in the highlights and shadows.

At the end of the day its about matching similar bits of images to other similar looking bits of images. What youā€™re describing sounds like goodĀ matches being found along one flight line, and good matches found along the adjacent line, but no matches between them. (You could check that out by picking a generated point and seeing what images it was matched in.)

Iā€™d give Lightroom a shot being careful not to lose data in the subject you are interested in. You could try adding a couple cross strips in the trouble spots. You could also try the geometric validation toggle, or even to increasing the number of required matches to be considered a valid point. And check your camera parameters, tie points and GCP!

As a last resort I was thinking you could use the annotation tool to mask out some areas of photos producing bad results, but understanding the nature of your problem more fully now I donā€™t like the idea as the block is no good if it isnā€™t tying between strips.

Let us know how you do!

Steve

Ā 

Ā 

Steve Yes, 7m. With regards to GCPs, true, but the thing is we were mapping really a large area 133sq.km, the drone I used then required many many flights. hence it was not practical installing GCP in every flight block, juts not cost effective. And this was not taken using wide lens, it was eBee which carries Sony WX, really something must be going on within the software. Due to this problem, I had to repeat flight with a bigger drone varying Sony A5100 with Voitglander. Things worked out, bu the voitglander images were much better resolution and quality.

Clarence

Clarence, did you try tying the strips together with some manual tie points?Ā Ok, too big an area to cover everything with GCP. But the manual tie points could help ensure one uniform block. Sounds like those bad blocks might only be holding the ABGPSā€¦

Most problems I have seen have had a good explanation. There are so many variables. But for several years now I have seen very robust triangulation from the software. It is always possible something changes or there is a new setting that catches us. But so far if I have everything set up properly and use good procedures I have gotten results that test very good against any photogrammetric program I have compared it with.

Steve

I hear you Steve and this what the support guys keep telling me and I have always told them yes MTPs are good and work nice for when you can easily apply them. But when you capture data in un-built areas like in my many cases, it is hard to find objects to use for marking MTPs. See for example link below; how do you easily use MTPs on the images of an area such as this? Yes I agree MTPs can dramatically improve your processing, but I have not found them easy to implement. When you have project where I can easily implement MTP the software already is able to process the project without issues, so no need for MPTs

https://www.dropbox.com/sh/g5ktcat4hacu5gf/AADiZiaJJTXDAK5R6KkcCdzra?dl=0

I know Pix4DNapper works differently, but just as an idea, if the stitching was not relying on point clouds/tie points alone, in addition to that the software could have been testing conventional stitching as it is done by Hugging or Microsoft ICE and use it to actually find matching pixels when conventional automatic matching fails, would this not be useful add-on? I am saying so because I have several times found for instance in some projects where I know I have good overlap pretty tight, and PIX4D wonā€™t calibrate, may be as you say the images could be dissimilar; but unless exposure is really difference between consecutive imagesĀ  Iā€™m still finding it hard to believe photos that in my eyes look similar the software would see them as dissimilar and hence fail to calibrate. I have to apologize, Iā€™m a little off topic but may be useful stuff to talk about. What I have not done so far is to compare different software packages with same set of data, may be that will say a lot more about Pix4Dmapper. make n mistake Pix4Dmapper is a great app.

Clarence

Hi Clarence, I looked at the one image that you posted. I will offer an opinion, but please know it is my own and that the Pix4D team may think differently. It is not an official opinion, ok?!

The image is showing very poor resolving power away from the center and toward the edges. Very blurry. So technically there may be a lot of overlap between flight lines but I donā€™t think it will be very useful in this case. The jpg compression is not helping in a scene like this either. The subject has a lot of similar colors and shading.Ā Ā 

I downloaded the image. This is a Sony A5100? But I see mixed reviews for the Sony 16mm prime lens, and technically some poor performance at wide-open apertures. You can lock the focus?

I think if this one photo is representative of the others there may be a lot of room for improvement. You might try smaller f-stopĀ to optimize the images for sharpness out to the corners and use less aggressive jpg compression. Might need to raise ISO a little to keep a fast shutter speed. Not a fantastic lens may need to experiment.

Steve

Ā 

1 Like

Hi Steve

I really like your comments and it is very true that the images focus fades away as you move towards the edges; as a result when you process the data, you end up these weird bulging up and down as shown in my results again placed in the same folder also discussed by agribox at the link below see and the image referenced in the next link. My similar outputĀ  saved in the same Dropbox location third link below.

http://agribotix.com/vignetting-effects-illustrated/

http://agribotix.com/wp-content/uploads/2016/03/VE3D_002.jpg

https://www.dropbox.com/sh/g5ktcat4hacu5gf/AADiZiaJJTXDAK5R6KkcCdzra?dl=0

Now, what makes me unhappy is that, every time you contact Pix4D team with issues such as this they will tell you either overlap or manual tie points. While in fact it is not just all image images overlapping, the quality of overlaps does not seems to gain any attention. Which I now agree with you that overlap alone is not enough, here needs to be good quality overlaps.

The Camera I used for this capture is surely Sony A5100 but the lens we used is Voitglander Aspherical II which is a 15mmĀ  and photos were taken in manual mode with the following parameters.

  • Side and Front Overlap both at 60%
  • Lens Setting: Infinity Focus, F5.6
  • ISO: 100
  • Shutter Speed:Ā  1/1250
  • Flight altitude 250m AGL
  • Aircraft speed between 12 and 17m/s

I continued testing with changed parameters, not much improvements as I think I did not find sweet spot, so I finally decided to switch to Aperture priority, which then gave significant improvement but also side overlap was increased to 75% front overlap to 70%. The results of which I can only share with you privately (in PDF) as it is clients final data, if you want to see let me know. But general comment is that there were great improvements even though the flights were done in different days, but excellent color balancing. The day before cloud and finally we had rain, the next day bright blue sky.

Thanks for your inputs. But quick one what manual settings would you use in a bright day and in relatively lower light/cloudy day?

Thanks for your input once again.

Clarence

Ā 

Clarence, I havenā€™t forgotten about responding. Just got dragged in a few too many directions. I have some more thoughts Iā€™ll scribble very soon.

Steve

Hi Steve

Ā 

Do not worry at all, I totally understand our primary businesses first, this forum is just to share knowledge and I appreciate your commitment.

Ā 

Thanks