Terrain Aware or Fixed altitude for Mapping (Ortho, 3D and cut/fill calculation)

[adm_geomatics]

Initial comments, but we're going to look at some data to see what's really what.

"Terrain awareness eliminates the variance in the distance to the ground and thus eliminates the discrepancies in overlap throughout sites with high topographic relief. Consistent overlap leads to more reliable data. From more consistent overlap and Ground Sampling Distance more points can be extracted and the orthomosaic resolution will be more uniform."

 

[yarrr]

I will contact an engineer and see what I can get, but you have to think about it with images per pixel and how that pixel looks. When tie-points are found they are matched and if an object looks different to the algorithm then it may not match it. This is the whole point of computer vision. Secondly if those points are different distances and thus angles that the point may not be seen combined with the matching problem then you end up with fewer tie-points for the reconstruction which means that you are missing possibly important features in between the points in the cloud. You are correct that the more intense the slopes the more important this becomes, but then that is the point of terrain following.

That generally makes sense, although one could make a counter argument that if you were flying over a slope, you would have more key points available to tie if the mission was flown at a constant altitude.

Maybe a better way to think about all of this is "What is the optimal altitude RANGE for SfM?" Then, if the survey range of the mission was out of the optimal RANGE, varying the flight altitude could be planned accordingly. This likely comes down to GSD, and if so, one would just plan a mission around the target for the lowest anticipated elevation.

Again, terrain following totally makes sense for areas with large relief, but I can't see a real justification for it on 'flatter' surveys.

 

[adm_geomatics]

you would have more key points available to tie if the mission was flown at a constant altitude

Actually no, because a slope distance flying with it is always longer than a horizontal distance.

What is the optimal altitude RANGE for SfM?

The optimal range is the time you are willing to spend mapping, the algorithm and the amount of computing power you have. I.E. I could take 20 hours mapping 10 acres with every photo equaling 1200sqft (instead of 12K), 60-70% overlaps, had the sufficient computing power and the algorithm tuned for that then I could make a virtually realistic 1:1 model.

For our use-case the optimal range is between 175 - 240ft (0.6 - 0.9in/px) depending on the size of the project and the desired type of output.

I can't see a real justification for it on 'flatter' surveys.

Agreed, but it would have to be VERY flat to not benefit. I mean, what's the downside? A square 10 acre tract at 1% fall would vary almost 7ft...

 

[yarrr]

Agreed, but it would have to be VERY flat to not benefit. I mean, what's the downside? A square 10 acre tract at 1% fall would vary almost 7ft...

I think the downside is complicating something that doesn't need to be complicated. Perhaps this would make more sense with a LiDAR payload, but I am still skeptical of any actual/meaningful improvements to a SfM point cloud.

Another reason for my skepticism is the resolution of the DEM data used for mission planning that people have mentioned above- on your theoretical 10 acre site- does it really make sense to try to match your UAV GNSS (accurate to 5m vertical), to a DEM (with a 10m grid size), for 7' of relief?

So, if anyone can point to literature/documentation showing a real reason to do this, please share. Really just curious if it is worth the effort.

 

[adm_geomatics]

complicating something that doesn't need to be complicated

I don't follow what you think is complicated?

actual/meaningful improvements to a SfM point cloud.

I'm not sure what you consider meaningful, but perfection with the acceptance of the limits of the technology is what we intend. Can we get from 2cm to 1cm consistently? Not without digging deeper and doing things like this.

on your theoretical 10 acre site- does it really make sense to try to match your UAV GNSS (accurate to 5m vertical), to a DEM (with a 10m grid size), for 7' of relief?

Yes, and that was way conservative. Many of us fly much more than 10ac at a time and a "flat" tract in Texas farmland can easily have 20ft of fall. Modern GNSS hardware is more like 1-3m, but most aircraft don't use GPS for vertical. More accurately they use the barometer and in the case of DJI including the vision sensors which is closer to 1m accurate most of the time. That's why the takeoff point in the brain of the system is zero and calculated up and not the actual HAE or MSL value.

literature/documentation showing a real reason to do this, please share.

Highly unlikely that there is good reliable data already available. Mapping with TA isn't as common as you would think and is a fairly new feature to this use in the grand scheme of things. Official documentation doesn't mean squat without numbers to back it up. I still don't see what the downside would be if there software is doing it by the simple input of three numerals. I'll test if you will?

 

[yarrr]

Highly unlikely that there is good reliable data already available. Mapping with TA isn't as common as you would think and is a fairly new feature to this use in the grand scheme of things. Official documentation doesn't mean squat without numbers to back it up. I still don't see what the downside would be if there software is doing it by the simple input of three numerals. I'll test if you will?

I'm with you!

Dug a little bit into Pix4d, and while they recommend TA, if you interpret this statement (and supporting equations) in the context of TA, we have healthy range to work with:

"Considering that the flight height and the GSD have a linear relationship, for the same project, captured with the same camera, the highest flight height at which images are taken should not exceed two times the lowest flight height"

My sites are usually shorelines or river corridors, often with under 20 feet of total ground relief- not worth screwing around with different flight altitudes.

However, I've got some potential bluff surveys coming up with closer to 150' total relief, so I'm trying to figure out if this is worthwhile. Processing similar projects in the past with Pix4d, I always seem to have a good (excessive?) number of match points. And that's kind of the point- there is a limited number of points in the images that are actually matched and used in the reconstruction. More data is not necessarily better. Even Pix4ds recommendations don't provide justification for a more accurate point cloud, just the ortho (via GSD).

For what it's worth, I also target something around a .8" GSD on all projects- which provides our team a high level of detail and accuracy at a file size that is usable in CAD/GIS. Recently chatted with an 'expert' who was proud that he flys his Inspire at something like 50' for a 1/4" GSD. Good luck actually using that data, for anything, at all.?‍

 

[adm_geomatics]

Cool one of my friends from DroneDeploy may hop on at some point to give us a little of their findings.

"Considering that the flight height and the GSD have a linear relationship, for the same project, captured with the same camera, the highest flight height at which images are taken should not exceed two times the lowest flight height"

This makes sense, but I would never let it separate that far unless I was trying to do a nice structure model and was taking pictures on the ground as well. This would apply for all drone imagery in the air so I guess they are saying if I fly at 150ft I should not let the ground get 300ft away from the drone. Our typical highest relief is about 200ft so this definitely comes into play and I remember clearly from a while back the first time I saw this problem when I flew from the high point of a steep road at 250ft and when the ground got 400ft from me the images looked horrible.

More data is not necessarily better.

That's the truth and there is such a thing as too much overlap before you start introducing noise into the algorithm. I use to run 75% overlaps until I was told this by a Photogrammetrist and that 60% sidelap was fine so I changed my standard config to 80/65. The other thing to think about is that with the aspect ratio being wider the distortion gets worse. Here's an example from a DroneDeploy report. This comes up sometimes when people want a 3:2 ratio over the 4:3 and while it's not a huge difference it does cause more distortion, but is also is not a huge difference in quality while using more data.

Recently chatted with an 'expert' who was proud that he flys his Inspire at something like 50' for a 1/4" GSD. Good luck actually using that data, for anything, at all.

Lol. Someone asked me the other day how they get to 0.05" because they wanted an ultra-realistic VR experience. It was an interesting use-case though.

 

[adm_geomatics]

Additional comments as received from the DroneDeploy Team.

"If used correctly, we'd expect terrain awareness to enable you to maintain a lower average altitude AGL for the same number of images captured.
Therefore you should expect higher resolution images of the ground on average, and so higher accuracy maps on average (due to more precise and accurate geolocation).

So...
1. You'll have higher resolution images of your GCPs and checkpoints, and so have the opportunity to tag them more precisely
2. The depth maps will be "higher resolution" for a given object on the ground
3. The point cloud will be higher density on the ground given the higher resolution depth maps of the ground
4. The mesh will be more precise, and higher fidelity: fine details will be preserved due to 3 and 2
5. The DEM will be "sharper" more precise due to 4.

Clearly precision is not equivalent to accuracy, but given the same GCP collect and a careful operator, you should expect lower checkpoint error from a terrain aware collect.

A nice example ortho tile from our testing:"

 

[yarrr]

That is a compelling graphic. But- still disingenuous, because the results would be equivalent if one planned a mission for a specific GSD at the lowest expected elevation.

As Chasco pointed out earlier, this would lead to excessive overlap at the higher elevations.

I don't mean to discourage anyone from thoughtful mission planning, but I really don't see any evidence that TA leads to higher accuracy point clouds (cut/fill calculations).

On the other hand, if people are using Dronedeploy (and blindly trusting whatever behind the curtain magic happens once the photos are uploaded); they have laid out a good argument for using TA- because their automated mission planning and post processing benifits from it.

 

[adm_geomatics]

That is a compelling graphic. But- still disingenuous, because the results would be equivalent if one planned a mission for a specific GSD at the lowest expected elevation.

As Chasco pointed out earlier, this would lead to excessive overlap at the higher elevations.

I don't mean to discourage anyone from thoughtful mission planning, but I really don't see any evidence that TA leads to higher accuracy point clouds (cut/fill calculations).

On the other hand, if people are using Dronedeploy (and blindly trusting whatever behind the curtain magic happens once the photos are uploaded); they have laid out a good argument for using TA- because their automated mission planning and post processing benifits from it.

Looking at the evidence provided by the resolution increase should be a pretty simple indicator of the accuracy that will occur through correct processing.

Linear flight produces insufficient overlap at high areas and excessive overlap in low areas.

We are about to conduct a formal testing program of DroneDeploy's TA and processing engine. We will shoot for 5 sites collected in both the linear and terrain following methods and full GCP and checkpoint analysis will be performed. Cut and Fill calculations will also be performed in several categories of that function. I am not sure by what is meant by magic behind the curtain unless it just refers to a process that only professionals understand.

 

[R Martin]

That is a compelling graphic. But- still disingenuous, because the results would be equivalent if one planned a mission for a specific GSD at the lowest expected elevation.

As Chasco pointed out earlier, this would lead to excessive overlap at the higher elevations.

I don't mean to discourage anyone from thoughtful mission planning, but I really don't see any evidence that TA leads to higher accuracy point clouds (cut/fill calculations).

On the other hand, if people are using Dronedeploy (and blindly trusting whatever behind the curtain magic happens once the photos are uploaded); they have laid out a good argument for using TA- because their automated mission planning and post processing benifits from it.

People that are serious about mapping and do it professionally are not generally using cloud-based apps to do the work for them. Can I point you to academic papers that prove to your satisfaction that TA mapping produces a sound product? No. Does flying a specific profile at a set altitude make a difference? I have noticed that it does. Of course my camera is probably a little bit more than what you are using. A 0.5GSD on an RX1RII Sony gets the job done at 400 feet AGL with minimal time spent on-site (the back end is another story). I map 100-150 acres in about 30 minutes with those settings. Try the same area with an Inspire (you'll be limited to about 100 feet AGL to hit the same GSD and hope you have a lot of batteries and time to do it).

What are the real benefits? I am comparing 1/2" pixels to 1/2" pixels instead of a hobknob collection of varying sized pixels. Does it matter? I digitize blocks of real estate without spending weeks or months in the field with the same or maybe a littler better accuracy than my GEO 7X GPS units. I spot check all the same. I rarely have to adjust. If you are really interested in the nuts and bolts talk to a surveyor. See how they do it and if it matters. Most of the shops we are in contact with that survey are pretty meticulous on how they gather and process data. Accuracy matters.

 

[adm_geomatics]

People that are serious about mapping and do it professionally are not generally using cloud-based apps to do the work for them.

I will respectfully disagree on this point. DroneDeploy and Propeller both provide results that are 95% of what can be done in a local manual process and a Pro would be that the data is much more efficient for business workflows. A Con for cloud processing is that you don't have as much control of the settings used for processing, but that is of little importance when the data is collected correctly. From a Surveyor's point of view.

Try the same area with an Inspire (you'll be limited to about 100 feet AGL to hit the same GSD and hope you have a lot of batteries and time to do it).

A Phantom 4 Pro is 150ft AGL for 0.5in/px and an Inspire with an X7 and appropriate 24mm lense would be 160ft. The RX1RII has 30-degree less FOV which means that the additional height of ground is not as advantageous as it initially appears. Unless you are using a different lens? What drone are you using and what overlaps? I would say though that with the narrower FOV you will not get as much horizontal distortion, but if your are using 3:2 you are introducing a little of it back in. Accuracy, every little bit of it, matters! It's a simple fact that as you approach perfection within limits the advances get smaller...

 

[R Martin]

I will respectfully disagree on this point. DroneDeploy and Propeller both provide results that are 95% of what can be done in a local manual process and a Pro would be that the data is much more efficient for business workflows. A Con for cloud processing is that you don't have as much control of the settings used for processing, but that is of little importance when the data is collected correctly. From a Surveyor's point of view.


A Phantom 4 Pro is 150ft AGL for 0.5in/px and an Inspire with an X7 and appropriate 24mm lense would be 160ft. The RX1RII has 30-degree less FOV which means that the additional height of ground is not as advantageous as it initially appears. Unless you are using a different lens? What drone are you using and what overlaps? I would say though that with the narrower FOV you will not get as much horizontal distortion, but if your are using 3:2 you are introducing a little of it back in. Accuracy, every little bit of it, matters! It's a simple fact that as you approach perfection within limits the advances get smaller...

Propeller, yes. I use Aeropoints as my control and have found them to be extremely accurate. DroneDeploy, maybe. I'm old and I don't trust what I can't see. I do my own processing via Pix4D.

Firefly 6 Pro flying 75% overlaps front and side. Proprietary flight control software. Based upon field measurements independently verified using mapping grade Trimble Geo 7X GPS units, I'm doing something right. It might not work for you. I does for me. We operate between mapping grade and survey grade as far as data quality. We aren't surveyors but we get as close as we can for a greatly reduced cost.

 

[adm_geomatics]

Propeller, yes. I use Aeropoints as my control and have found them to be extremely accurate. DroneDeploy, maybe. I'm old and I don't trust what I can't see. I do my own processing via Pix4D.

Firefly 6 Pro flying 75% overlaps front and side. Proprietary flight control software. Based upon field measurements independently verified using mapping grade Trimble Geo 7X GPS units, I'm doing something right. It might not work for you. I does for me. We operate between mapping grade and survey grade as far as data quality. We aren't surveyors but we get as close as we can for a greatly reduced cost.

I didn't mean to sound like I didn't trust your results or methods so apologies if that is how it came across. i think our statements together prove that there is more than one way to skin a map. That said though, how are you getting centimeter accuracy with a Geo? When you say mapping grade are you trying to say survey? Mapping grade to me is GIS. It may be off-topic, but I think it may help allot of people that don't survey or know GNSS.

 

[R Martin]

I didn't mean to sound like I didn't trust your results or methods so apologies if that is how it came across. i think our statements together prove that there is more than one way to skin a map. That said though, how are you getting centimeter accuracy with a Geo? When you say mapping grade are you trying to say survey? Mapping grade to me is GIS. It may be off-topic, but I think it may help allot of people that don't survey or know GNSS.

The Geo 7X is a GPS unit. When you order one you are asked for the accuracy that the unit will provide and then Trimble loads that software package into the unit. They all have the ability to be survey-grade accuracy if you are will to pay the cost. We chose, because we are not surveyors, to have the mapping-grade package loaded. Under the right conditions with the GPS god aligned perfectly with the cosmos. we get 2 centimeter accuracy. The Aeropoints, on the other hand, return survey-grade accuracy in the millimeter range. With them deployed properly and no multipathing or other signal degradation, I usually record under 10 millimeters on the X/Y plane. Height is another matter altogether. GPS inaccuracy in the Z plane can vary.

GIS can be as accurate as the data you put into it. That is a completely different ball of wax. We combine various grades of accuracy and grade each bit of data based upon its accuracy (SUE level A is as good as it gets; SUE level D is word of mouth not that reliable). Things that matter (basemap of pervious and non-pervious surfaces, buildings, land, parcels, floor plans, surface and sub-surface utilities, etc.) are recorded at the highest level of accuracy that we have and are constantly updated as that changes. Trees, outdoor furniture, signage (to name a few) do not require that high level of accuracy so they are tracked at a lower level. One university up north has a 30+ person GIS department. They have two people on staff that do nothing but track outdoor furniture. Our department has three full-time people and a few more students. You do what you can do with what you got. We don't place a lot of emphasis on non-critical stuff, but we do track non-critical items to a degree.

UNT GIS Open Data page
Click on the first item on the page for the COR report. Zoom in to the basemap. Viola! Magical pig. And the trees are not there for show. They really are there; all 5000+ at last count.

 

[adm_geomatics]

I promise, the last hijack here...

So the Geo is more for locating other objects and maybe some checkpoints?

How long does it take for the Aeropoints to collect a mm-level coordinate? My experience is that they are 1cm horizontal and 2cm vertical on average. How do you reset them when you return to a site and deal with surface coordinates?

 

[yarrr]

We chose, because we are not surveyors, to have the mapping-grade package loaded. Under the right conditions with the GPS god aligned perfectly with the cosmos. we get 2 centimeter accuracy. The Aeropoints, on the other hand, return survey-grade accuracy in the millimeter range. With them deployed properly and no multipathing or other signal degradation, I usually record under 10 millimeters on the X/Y plane. Height is another matter altogether. GPS inaccuracy in the Z plane can vary.

This is exactly why I don't believe that TA is worth the hype!!!

I also don't want to make this an equipment contest, and I'm not criticizing your setup.

But... How could you possibly know if your TA processed dataset was more accurate than your non-TA when you are verifying with that equipment? Out of curisoity, do you have RTK/PPK on the Firefly?

 

[adm_geomatics]

This is exactly why I don't believe that TA is worth the hype!!!

I also don't want to make this an equipment contest, and I'm not criticizing your setup.

But... How could you possibly know if your TA processed dataset was more accurate than your non-TA when you are verifying with that equipment? Out of curisoity, do you have RTK/PPK on the Firefly?

Their equipment has nothing to do with TA and that is where we got off on a tangent when equipment came into the discussion. If anything you had the right question about RTK/PPK on the UAV. Testing will include native GNSS and augmented via Emlid receivers. You can buy a multipath base and rover setup for less than what the Geo cost and you don't need a subscription. Base, rover, mobile receiver, antenna and the P4P cost us $4k. That is sub-centimeter accuracy and on a localized network so we are on perfect coordinates in relation to the project sites every time. This all leads to 2-3cm stakeout so my pie-in-the-sky is 2cm consistently with TA.

 

[R Martin]

I promise, the last hijack here...

So the Geo is more for locating other objects and maybe some checkpoints?

How long does it take for the Aeropoints to collect a mm-level coordinate? My experience is that they are 1cm horizontal and 2cm vertical on average. How do you reset them when you return to a site and deal with surface coordinates?

The Geo is just another tool in the arsenal. Unfortunately, I am in a dense urban setting with tons of foot traffic and I am restricted on what I can fly and still abide by the regs. So when I can't fly it, I use the Geo to collect accurate GPS data on utilities (specifically) and other features of interest. Currently, I am validating our geodatabase street records. I can't fly it due to staffing (I'm the only one in the office and the shelter in place does not go into effect until midnight tonight). So the Geo is the next best tool we have to collect point data.

The minimum time that Aeropoints need to be operating is 45 minutes. We set control first and then set the base station up. After that we set the base up and then fly the mission. After tearing and packing everything the Aeropoints have been running about 3+ hours. Side note: they automatically shut down at around 4-4.5 hours. When we get within range of the network they upload their data. I process via the portal and then grab the numbers and plug that into Pix4D prior to stitching everything together. If I remember correctly they run in WGS84 native. We have that exported to NAD83(2011) State Plane 4202. We are lucky enough to have a CORS site operated by TXDOT less than a mile away.

Each flight is unique so we don't have fixed placements. Part of our Title One document for new construction sets a requirement for the general contractor to set a minimum of four surveyed benchmarks for each new construction project. We specify the locations of the monuments and then use those to set up a GPS base station that links directly to the aircraft to provide an RTK connection stream to improve the GPS picture.

 

[R Martin]

This is exactly why I don't believe that TA is worth the hype!!!

I also don't want to make this an equipment contest, and I'm not criticizing your setup.

But... How could you possibly know if your TA processed dataset was more accurate than your non-TA when you are verifying with that equipment? Out of curisoity, do you have RTK/PPK on the Firefly?

Yes. We set up a base station on a surveyed benchmark that provides an RTK stream to refine the aircraft's GPS numbers. I've flown both ways; both with an RTK stream supplemented with Aeropoints as checkpoints as well as without an RTK stream using only the Aeropoints as control points and checkpoints. I've flown with and without terrain avoidance ( and that term is not accurate in this case) which only became available on the FireFly sometime last summer?? It makes a difference.

I know by looking at the data and comparing it, then spot checking it in the field. I don't take anything for granted. I'm old school. Technology breaks when you need to rely on it the most. I don't trust anything I haven't field verified when it comes to the important stuff. A lot of things we collect fall outside that level of scrutiny. It doesn't matter if a tree is exactly at that given location of if its 5 inches east of where we drop the point. OTOH, a 13.2 KVA duct bank is down to the centimeter, as is a fiber trunk line. Some things you can't afford to find accidently. At that voltage everyone about 8 feet around the excavator will be a mort.

The bottom line is you have to trust the equipment. But I don't trust the equipment so I have more equipment to check the numbers multiple times. If they come up close (1-2cm) THEN I know we did it right.