Terrain Mapping One Square mile ?

[LBD511]

I have a Mapping mission the will result in thousands of photos total. Normally I use WebODM which I have running in Docker on my Mac, the most photos I have processed was 1200 and that took a long time, I cannot afford Pix4D or any that charge for credits at this time.

this is a list I will look into: are there any others?

• DroneDeploy 3D mapping mobile app
• Pix4D Mapper photogrammetry
• DroneDeploy Enterprise 3D Map
• AutoDesk ReCap photogrammetry
• SimActive Correlator3D™ software
• Maps Made Easy orthophoto and 3D models
• 3DF Zephyr photogrammetry software
• Agisoft PhotoScan photogrammetry
• PrecisionHawk 3D map software
• Open Drone Map photogrammetry
• ESRI Drone2Map for ArcGIS
• Agisoft Metashape 3D software

The project is Terrain mapping one square mile I have worked out the mission using Good Earth and my M30T RC, I have done 40 acre parcels they take about 4 hrs. drone time and average 6000 photos I have not processed these yet, it will take 16 of the 40-acre parcels to make the square mile, so that's an option.

Has anyone done any mapping of this size? (about 484 football fields)

A square mile is 640 acres and if I max the detail specs with Terrain following using 83% overlap and all 5 camera methods, set AGL to Terrain follow at 50.3m, (sorry not using the correct technical lingo) I get the following: configuration for the mission:

649-acre area
GSD 1.79cm/pixel
Oblique GSD 2.53cm/pixel
Distance 1093.21km
Est Duration 52 hours (40 Acre parcel tales about 4 hrs.)
Photos 102992

Pretty awesome challenge in that it will take several days, lots of battery swapping, not sure it is even possible given the M30T only has a 128GB Simm .

I will do it, not sure how I can break it down into smaller segments. thoughts?

Note this POST was edited to correct grammar (what I could fine anyway) and mixing up Meters vs Feet My apologies for the confusion.

Phil

 

[adm_geomatics]

This is quite the task. Just to start a discussion let's make sure I am understanding the parameters. You are trying to map 650ac at 50ft AGL? Using the real-time terrain following? Is there a reason you need to fly so low?

 

[LBD511]

There are no roads there yet need to find and mark (log/Lat) for creek beds, cattle trails and rocks and sage brush, hills, big holes etc., potential well drilling require details to plan best access, there are no trees or buildings.

Also need to pick up dirt bike signs if any and snake/critter habitats to be as careful about the environment as possible.

However, I realize I can do 100,200,300,400ft runs "Drone height" (not Terrian follow distance), but the goal is to know what it takes to get the most detail, these higher views are more than likely going to be done anyway. BTW the default Terrain Follow value when setting this up was 32.5m (not ft as stated before thanks Adm_geomatics for the UI snapshot catch)

So, the 35.3m not 50ft measurement is the ultimate goal. (this statement was corrected to show actual values I mixed up meters and feet previously)

I have been doing 120m-150m (not ft) terrain follow near my home and its ok but, I just want to see what's the differences. At home I have trees and buildings to consider.

The project will also need a nice video covering the entire parcel for marketing purposes, I need to experiment the best height for that too, showing possible access routes and building sites.

The plan is to Park my 12x18ft trailer on the land ( I cannot drive at Night) and stay until all the variations are captured.

Its off a dirt road 5 miles from a main paved road and 30 minutes from a local town.

Hope that helps and not too much information.

Note : these comments were edited to make corrections where feet were mistakenly used instead of meters, my apologies

 

[Dave Pitman]

You are certainly free to do as you please. But 50 foot AGL for typical terrestrial mapping is way too low. Obstacles are not the reason that higher is better. If your primary goal is to catalog 2" tall vegetation, then maybe.

 

[LBD511]

hmm 2-foot-tall vegetation is more like it, wonder why the default is so low? thought there would be a reason.
What is the recommended "typical" height in your opinion?

I can search around for other demos and training; not sure they would be consistent.

Perhaps after doing a 40-acre parcel or two and try different heights as I go I'll have a better idea.
But the downside is I cannot process the files on location, no reason except time and gas I cannot make more than one trip.
It is a five-hour drive each way to the site.

Attached the default Terrain Altitude "35.3m follow of a 40-acre parcel after camera, AGL selection, I increased to 50.3m for no particular reason except I did some work at 120 ft and wanted to go lower, but thought 35.3m was a bit too low. (Edited to correct measurements incorrectly stated previously)

Note the default overlap was 70 I changed it to 83 per recommendations from multiple sample, demos and training videos.

 

[adm_geomatics]

There are no roads there yet need to find and mark (log/Lat) for creak beds, cattle trails and rocks and sage brush, hills, big holes etc., potential well drilling require details to plan best access, there are no trees or buildings.

Also need to pickup dirt bike signs if any and snake/critter habitats to be as careful about the environment as possible.

However I realize I can do 100,200,300,400 foot runs, but the goal is to know what it takes to get the most detail, these higher views are more than likely going to be done anyway. BTW the default when setting this up was 32.5 FT

So the 50ft measurement is the ultimate goal.

I have been doing 120-150 ft terrain follow near my home and its ok but, I just want to see what's the differences. At home I have trees and buildings to consider.

The project will also need a nice video covering the entire parcel for marketing purposes, I need to experiment the best height for that too, showing possible access routes and building sites.

The plan is to Park my 12x18ft trailer on the land ( I cannot drive at Night) and stay until all the variations are captured.

Its off a dirt road 5 miles from a main paved road and 30 minutes from a local town.

Hope that helps and not too much information.

Are you mapping it with thermal? How do you expect to identify animal habitats? As for trail signs, they may be visible in the images but it's doubtful they will render in the model and may not even show up in an orthomosaic depending on the frames that the software selects for that exact location.

That said something isn't jiving with your numbers. I just did a test mission in FlightHub 2 and it wouldn't even let me set the TF AGL below 82ft (25m) and that was 0.89cm/px. At 117ft AGL it reports 1.27cm/px or 0.5in/px which is very low for most mapping missions. I can't even get my UI to refresh and show the flight lines at 82ft...

 

[adm_geomatics]

hmm 2-foot-tall vegetation is more like it, wonder why the default is so low? thought there would be a reason.
What is the recommended "typical" height in your opinion?

I can search around for other demos and training; not sure they would be consistent.

Perhaps after doing a 40-acre parcel or two and try different heights as I go I'll have a better idea.
But the downside is I cannot process the files on location, no reason except time and gas I cannot make more than one trip.
It is a five-hour drive each way to the site.

Attached the default Terrain Altitude "35.3 follow of a 40-acre parcel after camera, AGL selection, I increased to 50.3 for no particular reason except I did some work at 120 ft and wanted to go lower, but thought 35.3 ft was a bit too low.

Note the default overlap was 70 I changed it to 83 per recommendations from multiple sample, demos and training videos.

Also note that your UI is in metric, not feet.

 

[LBD511]

Are you mapping it with thermal? How do you expect to identify animal habitats? As for trail signs, they may be visible in the images but it's doubtful they will render in the model and may not even show up in an orthomosaic depending on the frames that the software selects for that exact location.

That said something isn't jiving with your numbers. I just did a test mission in FlightHub 2 and it wouldn't even let me set the TF AGL below 82ft (25m) and that was 0.89cm/px. At 117ft AGL it reports 1.27cm/px or 0.5in/px which is very low for most mapping missions. I can't even get my UI to refresh and show the flight lines at 82ft...

thanks I just took the Google-Earth KML files uploaded to the M30T RC , updated the settings to Oblique collections, altitude mode AGL , uploaded DSM file from network, set Terrain follow to 50.3m from 35.3m, set all overlaps to 83, left all others to default here is a snapshot of anther 40 acres I just did exactly the same.

SO your saying there is something wrong with the values provided by the RC ?

 

[LBD511]

Also note that your UI is in metric, not feet.

sorry your correct 35.3 m is the default see there is a reason I post this stuff, thanks, I am probably the only idiot that makes that mistake, and I have been doing this a whole 6 months on weekends.

so, default is 115.8136 feet which makes more sense, changing the value to 53.3 is 165.02 feet. looks like will go back to the default.

 

[LBD511]

thanks I just took the Google-Earth KML files uploaded to the M30T RC , updated the settings to Oblique collections, altitude mode AGL , uploaded DSM file from network, set Terrain follow to 50.3 m from 35.3 m, set all overlaps to 83, left all others to default here is a snapshot of anther 40 acres I just did exactly the same.

SO your saying there is something wrong with the values provided by the RC ?

well, I am learning, yes, I will be testing the Thermal also, I don't know how else to learn but practice and to just do it, it has worked so far at my age anyway. my error the ft values are corrected to m where I could see the errors, thanks for the feedback.

 

[adm_geomatics]

Mostly pointing out that your interface is reporting metric but your figures are based on feet. Also, your original post reported 1.79cm/px which would be much higher than the 117ft reporting 1.27cm/px I got. Is that maybe reporting the IR sensor?

 

[LBD511]

I will check the first snapshot was the 649 acres I just reset it to default 35.3m for Terrain Follow, it takes an hour or so to update the route, that was a shocker! to say the least. When it is done, I will check that value again. currently I see Ortho GSD 1.26cm/pixel , Oblique 1.77cm/pixel

 

[MapMaker53]

I cannot afford Pix4D or any that charge for credits at this time.

So many technical questions and answers here, but I have a very basic one. Why aren't you charging your client for the software subscription and the processing fees for this massive undertaking. Surely this can't be just a personal project that you are doing for fun. Just curious.

 

[LBD511]

So many technical questions and answers here, but I have a very basic one. Why aren't you charging your client for the software subscription and the processing fees for this massive undertaking. Surely this can't be just a personal project that you are doing for fun. Just curious.

it is my land to learn on, I have a potential client who wants a price to do 40-acre parcels, once I understand what it takes to do exceptional work then I will know what to charge, by the hour or per acre? just haven't figured it out, so far, its looking like $250/hour or $1500/40 acre parcel plus $.67/mile.

Birds eye view once mapped might be selling off a few 40 acre parcels to pay for common well drilling.

 

[adm_geomatics]

So many technical questions and answers here, but I have a very basic one. Why aren't you charging your client for the software subscription and the processing fees for this massive undertaking. Surely this can't be just a personal project that you are doing for fun. Just curious.

I agree. If you chose a subscription it should definitely be part of general conditions.

 

[adm_geomatics]

it is my land to learn on, I have a potential client who wants a price to do 40-acre parcels, once I understand what it takes to do exceptional work then I will know what to charge, by the hour or per acre? just haven't figured it out, so far, its looking like $250/hour or $1500/40 acre parcel plus $.67/mile.

Where are you located? If you are getting private deals that could work out well, but if you are soliciting it may be a bit high for your regional economy.

 

[jaja6009]

I knid of breezed through this thread, so sorry if this has been asked/answered.

If you process using your own computer, do you have a computer that can handle this?

With over 100,000 images will need a lot of RAM. I would ask the software support you choose if this is even advisable. You could split it into different sections and then combine the final project.
The 12 MP images from your Wide camera will help though since they are smaller than the 20MP Mavic 3E/Phantom 4 RTK.

I would love to see if Pix4Dmatic is up to the task since it is touted exactly for this type of large scale use case.

Also you can cut down on the overlap if you are using double grid oblique to try to cut down on images.

 

[adm_geomatics]

I knid of breezed through this thread, so sorry if this has been asked/answered.

If you process using your own computer, do you have a computer that can handle this?

With over 100,000 images will need a lot of RAM. I would ask the software support you choose if this is even advisable. You could split it into different sections and then combine the final project.
The 12 MP images from your Wide camera will help though since they are smaller than the 20MP Mavic 3E/Phantom 4 RTK.

I would love to see if Pix4Dmatic is up to the task since it is touted exactly for this type of large scale use case.

Also you can cut down on the overlap if you are using double grid oblique to try to cut down on images.

The only way to process this much data without a node system is to build separate chunks and combine them as you assumed. I’ve tried pretty much every processing software out there and Matic is no faster than anything else apples to apples. It is pre-configured to operate just like like we are talking about. I use Metashape and Matic was about 10% faster but the result were inferior when you dive into the accuracy of the photo alignments and point cloud confidence. What Metashape does way better is operate as a node system. You are better off with two $2000 machines than one $4000.

 

[adm_geomatics]

Long but informative read from my latest discussion with ChatGPT.

Your observation about Agisoft Metashape’s advantages in processing large datasets through chunking and distributed computing is valid and highlights a key difference between the two software solutions. Let’s break this down and assess your assumption about Agisoft Metashape’s superiority for distributed, large-scale processing:

1. Chunk-Based Processing in Metashape

Agisoft Metashape allows datasets to be divided into “chunks,” which can represent discrete sections of the project. This offers several benefits:

• Parallel Processing: Each chunk can be processed independently, enabling efficient use of multiple machines or resources in a distributed setup.

• Memory Management: By processing subsets of data, Metashape avoids overwhelming a single machine’s memory, making it better suited for extremely large datasets.

• Node System Flexibility: Using the Python API or batch processing, you can set up highly customized workflows to handle multiple chunks sequentially or concurrently. This mimics a “node system” where you control the dependencies and outputs of each step.

Pix4Dmatic lacks an equivalent “chunk-based” system, requiring the entire dataset to be processed in a monolithic workflow. This can lead to resource bottlenecks when dealing with vast datasets.

2. Distributed Processing in Metashape

Metashape’s support for network processing significantly enhances its scalability:

• Multiple Machine Support: With Metashape’s distributed processing, chunks or individual processing steps (e.g., dense cloud generation, orthomosaic creation) can be offloaded to multiple machines on a local network or cluster.

• Cost-Efficiency: Distributed processing reduces the need for a single, powerful workstation, spreading the load across less expensive hardware.

• Scalability for Ultra-Large Projects: This architecture excels in scenarios like country-wide aerial surveys or large infrastructure mapping, where datasets consist of tens or hundreds of thousands of images.

Pix4Dmatic, in contrast, does not natively support distributed processing. All processing happens on one machine, limiting scalability and requiring extremely powerful hardware for massive projects.

3. Pix4Dmatic’s Optimization Tradeoffs

While Pix4Dmatic’s architecture is not inherently designed for distributed processing, it is optimized for single-machine performance with large datasets. It may outperform Metashape in scenarios where:

• The dataset fits comfortably within the hardware’s limits.

• The workflow requires minimal manual intervention or chunk-based customization.

For large-scale distributed processing, however, Metashape’s chunking and network processing provide an undeniable advantage.

4. Practical Considerations

• When Metashape Excels: If your workflow involves massive datasets, requires distributed resources, or demands custom chunk-based processing, Metashape is superior.

• When Pix4Dmatic Excels: If you’re working with moderately large datasets and want an out-of-the-box solution with minimal setup, Pix4Dmatic’s faster, streamlined workflow can still be a good choice.

Conclusion

Your assumption is accurate: Agisoft Metashape’s chunking and distributed processing capabilities make it more versatile and faster for truly large-scale projects that can benefit from parallelism. However, Pix4Dmatic remains competitive in specific scenarios where processing is confined to a single machine, and the dataset is not so massive that chunking or distribution becomes necessary.

 

[jaja6009]

The only way to process this much data without a node system is to build separate chunks and combine them as you assumed. I’ve tried pretty much every processing software out there and Matic is no faster than anything else apples to apples. It is pre-configured to operate just like like we are talking about. I use Metashape and Matic was about 10% faster but the result were inferior when you dive into the accuracy of the photo alignments and point cloud confidence. What Metashape does way better is operate as a node system. You are better off with two $2000 machines than one $4000.

I also am a Metashape convert. Pix4Dmapper was only my first photogrammetry software due to the college I attended/worked at used it. Pix4D's marketing is top notch.

Once I tried out Agisoft and then started taking advantage of its no default settings and workflows, I never looked back.

I own a copy of Metashape Pro, and at work we have Pix4D matic, mapper and react. The other people using the software were wowed by Pix4D's marketing and insisted on it. There is some buyer's remorse though when comparing any models as Agisoft's models have gotten so good with software updates since I first starting using it. Pix4D absolutely sucks if there are power lines in your model, balloon city, which requires a bunch of manual editing in the point cloud and reprocessing to remove.

OP, please keep us updated on this project, as it has my interest.