How to do forest surveys

[GrinGEO]

I have a client request to do surveys in forest withouth NDVI due to the cost. The clients believes that he is able to see from the Foto wich trees are sick...

But he need a georef orthomosaic of a large area (30 hectares)

The problem is that Ortho does not produce clear images, but do a lot of distortion / warping effect especiially on trees....

Is there any of you proccing forest surveys and whats the best way to become as clear images as possible?

Thanks

[Jeremiah_ROWE]

Classify and filter out everything but the ground points. Create a DEM from just the ground class. Build the ortho from the ground-based DEM. You will have to manually assign photos in some areas to keep the "lean" consistent throughout the imagery area.

Are you going to be utilizing GCPs? Will this be going into the CAD environment? What UAV will you be using?

It really sounds like this is a better application for LiDAR combined with photogrammetry!

[SAV]

Hi GrinGEO,

You should probably fly a grid survey (second set of flight lines perpendicular to the first set) and capture oblique imagery (instead of nadir) to improve photogrammetric reconstruction results which should help with the distortion issues. Also use a high image overlap (80-90%).

You could also try a multi-camera setup to improve your data acquisition.

Additionally, if you don't need very high resolution imagery, aim to fly quite high above the trees. This will give you less detailed imagery, but it also helps to reduce distortion in your orthophotograph. Have a chat with your client to find the sweet spot where he is still able to tell if it's a 'sick tree' or not. A GSD of 10cm/pixel vs 3cm/pixel makes a big difference from a data acquisition and processing point of view, but might not make a big difference from a 'fit for purpose' point of view. If you can see what you want in the 10cm/pixel imagery, then there is no need to go down to 3cm/pixel.

Last but not least. Collect your data on days when there is no wind. Trees/branches/leaves are generally moving in the wind which is a big issue for photogrammetric reconstruction algorithms (i.e., SfM). Some of your distortion problems might be inherited this way.

All the best. Let me know if you were able to improve your results.

Regards
SAV

[dowallin]

SAV,
When you mention the use of oblique imagery or a dual camera system, could you elaborate?

Do you recommend having a single camera pointed a bit forward or off to one side? And with a dual camera system, would this involve cameras pointed off to either side or perhaps one pointed forward and the other at nadir?

I'd like to try this. I've done quite a bit of flying over forests with a great deal of overlap and often end up with gaps of unaligned images an sections with a large amount of distortion.

Are there any publications (books, scientific publications or online resources) that elaborate on this issue?

Thanks
David

[StevenKelly]

Classify and filter out everything but the ground points. Create a DEM from just the ground class. Build the ortho from the ground-based DEM. You will have to manually assign photos in some areas to keep the "lean" consistent throughout the imagery area.

Are you going to be utilizing GCPs? Will this be going into the CAD environment? What UAV will you be using?

It really sounds like this is a better application for LiDAR combined with photogrammetry!

Good to know about that.

[SAV]

Hi David,

Oblique imagery
Camera is (slightly) tilted and not facing straight down (= nadir). Don't tilt it too far though; you DO NOT want to include the horizon/sky. In your case (single camera setup) your camera should be pointing in the flight direction and be tilt down (about 60-70 deg from horizontal). Fly a 'standard' mission with parallel flight lines and then a second mission with is oriented perpendicular to the first mission.

Multi-camera setup
Camera rig with several cameras that are all oriented differently. For example, one oblique forward, one oblique backward, one oblique left, one oblique right and one nadir. All cameras would be triggered at the same time. More details: https://pix4d.com/ultra-efficient-photogrammetry-rig-processing/

Note that you can only get a good orthophotograph that is properly corrected for distortion if you have generated a high quality dense point cloud or DEM. The dense cloud's or DEM's elevation information (depending on what you have chosen in Photoscan) will be used to orthorectify your images. For example, if you have processed the DEM or dense point cloud at low quality settings, your orthophoto will not be properly orthorectified (even though it might be visually looking better).

As already mentioned, tree/branch/leave movements are an issue. By flying higher and/or at days with no wind these effects can be mitigated.

Regards
SAV

SAV,
When you mention the use of oblique imagery or a dual camera system, could you elaborate?

Do you recommend having a single camera pointed a bit forward or off to one side? And with a dual camera system, would this involve cameras pointed off to either side or perhaps one pointed forward and the other at nadir?

I'd like to try this. I've done quite a bit of flying over forests with a great deal of overlap and often end up with gaps of unaligned images an sections with a large amount of distortion.

Are there any publications (books, scientific publications or online resources) that elaborate on this issue?

Thanks
David

[GrinGEO]

The point is that all depends on costs, so it should be as cheap as possible and not go over 1500 EUR / 20 hektares, or it wont be interesting

I have no need for any dense cloud but only a very good Orthomosaic that is georef withouth any blur or warp artefacts casued by agisoft, so probably I will also not do any mosaic blending but have hard edges from the pictures....

Client only wants highres orthofoto to see where trees are sick and need to be cutted

[Jeremiah_ROWE]

Here is a quick example showing the results of a single nadir pass vs. double perpendicular grid at slightly oblique angle data collection methods.

The single nadir pass missed under the trees, the power pole completely, and the sides of the buildings are much weaker. I have highlighted the main areas of interest to see what I am referring to.

[dowallin]

SAV,
thanks for your response. I mostly use a 3DR Solo with a Canon S100 camera. I use Mission Planner for grid flights. I have several S100s; standard RGB and two modified for IR. At times, I've flown two nadir-pointing S100s at the same time (two different IR cameras). Yes, this is quite a load for the Solo but it flies just fine, albeit with a reduced flight time. It occurs to me that I could fly two RGB cameras, one nadir pointing and one pointing forward a bit to get oblique images. I took a look at the Pix4D document regarding "rig camera" and I get the issue of needing simultaneous camera triggering. With my S100s, I use the CHDK intervelometer so the cameras DO NOT trigger at the same time.

But I'm wondering if the set up that I've described, and processing the oblique and nadir images together, would produce good results? Or better to follow your suggestion of a single oblique camera with two perpendicular flight plans.

Also, I recognize that, when flying over tall forest canopies, the effective overlap at the top of the canopy is much less than the ground-based overlap that Mission Planner calculates (see attached). And, as you point out, canopy movement with the wind can make it very difficult to align images.

David 

[dowallin]

Jeremiah,
I love your comparison of oblique vs. nadir images. This may be a dumb question, but in the final ortho image, shouldn't the power pole just look like a dot? And shouldn't all buildings just appear as rectangles with the sides of the buildings NOT be visible?

David

[Jeremiah_ROWE]

Thanks, and it is not a dumb question. A more complete 3D reconstruction can help lead to a better DEM/mesh base for an ortho. It all depends on what you are trying accomplish. My purpose for these examples is going into the 3D world, not just a top-down stitched image.

[GrinGEO]

Thanks Jeremaya for your feedback, it helps a lot for urban mapping purposes.

In this case client is requesting only a fast Orthofoto process that cost as less as possible. Dense Cloud Generation is taking very lot of time, if we can find a process that takes less time for deliverey within 24h this would be great

[SAV]

Hi dowallin,

A true 'rig camera setup' allows you to trigger all cameras at the exact same time. Pix4D can actually use the geometrical information of the camera rig itself (distance between cameras, orientation of each camera) to improve/accelerate photogrammetric modelling. PhotoScan, however, does not have such camera rig support. Nevertheless, you can still process such datasets, even when the cameras are not synchronized as it is the case for your 3DR Solo setup.

Possibly one of the best setups would be 2 oblique cameras (pointing down at  about 70 degrees from horizontal, no horizon visible in images), one facing forward (in flight direction) and one facing backward flying a grid survey (two perpendicular flight plans).

As you mentioned correctly, image overlap in most flight planning apps assumes a flat surface. In reality, your image overlap (and also your ground sampling distance (GSD) is changing throughout the scene. Tall objects such as trees will have a smaller image overlap but also a smaller GSD (=higher resolution).

As mentioned earlier, if you choose a larger GSD (=lower resolution) the effects of trees/branches/leaves moving in the wind will be less 'visible' in pictures. For example, if your GSD is 10cm/pixel and a tree is moving by 10cm, then this is basically happening 'within' that pixel. However, if your GSD was 0.5cm/pixel, then a 10cm tree movement would be visible in 20 pixels. This would be a problem for SfM algorithms which try to detect and match features in overlapping images because and assume that the objects didn't move. Therefore, if you fly higher (higher GSD = lower resolution) you are minimizing the effect of moving trees. You need to find the right balance between best modelling results and resolution.

Regards
SAV

SAV,
thanks for your response. I mostly use a 3DR Solo with a Canon S100 camera. I use Mission Planner for grid flights. I have several S100s; standard RGB and two modified for IR. At times, I've flown two nadir-pointing S100s at the same time (two different IR cameras). Yes, this is quite a load for the Solo but it flies just fine, albeit with a reduced flight time. It occurs to me that I could fly two RGB cameras, one nadir pointing and one pointing forward a bit to get oblique images. I took a look at the Pix4D document regarding "rig camera" and I get the issue of needing simultaneous camera triggering. With my S100s, I use the CHDK intervelometer so the cameras DO NOT trigger at the same time.

But I'm wondering if the set up that I've described, and processing the oblique and nadir images together, would produce good results? Or better to follow your suggestion of a single oblique camera with two perpendicular flight plans.

Also, I recognize that, when flying over tall forest canopies, the effective overlap at the top of the canopy is much less than the ground-based overlap that Mission Planner calculates (see attached). And, as you point out, canopy movement with the wind can make it very difficult to align images.

David

[SAV]

Hi GrinGEO,

I think as long as you make it clear to your client that low cost does NOT give you the best quality, then that's OK.

However, your costumer still has the option to get better data from the same set of images if it is processed at higher quality settings. This will take more time and therefore cost more. No magic here 

Regarding delivery within 24h; this is probably depending on your hardware. More powerful hardware >>> quicker processing. But better hardware costs more, which probably influences the price of your final product.

Regards
SAV

Thanks Jeremaya for your feedback, it helps a lot for urban mapping purposes.

In this case client is requesting only a fast Orthofoto process that cost as less as possible. Dense Cloud Generation is taking very lot of time, if we can find a process that takes less time for deliverey within 24h this would be great

[GrinGEO]

Is there any way in the PS Raster Calculator to do some kind of False Color settings like this of a forest survey?