A totally wrong survey

r62ewa · May 23, 2017

Same here. The cost of an RTK unit is just too much at present to improve the absolute for me.

I tried to downgrade my firmware on my Phantom but there was no option in the DJI Assistant 2 software to downgrade as I've seen in other videos, just an option to Reset to factory settings, which I did not do.

PeteM · May 23, 2017

I repeated the flight (no CGPs and lower altitude).

I also performed a "classic" topographic survey.

In the next few days (as soon as I have time to process the data) I'll let you know.

PeteM · May 25, 2017

I, finally, compared the new topographic survey with drone (without GCPs) with a classic topographic survey (total station with high precision).

For a good 80% of the entire area to be mapped the measurement differences fall into acceptable error (max 10 cm), only in a small portion of the area - corresponding to the very lower part of the hole filled with flatty white material - errors start to rise up to about 60 cm... errors rise up even to a max of 1 m just just outside the lower border of the hole...

In the lateral area, next to the one where errors are found, far from the white hole, measures are again correct.

The orientation along z is correct. The error is only in the value of the measurement.

What happen ? How can I avoid it in the future ?

Richard R · May 25, 2017

Sounds like you've pretty much answered your own question. That flat featureless surface is messing with the algorithm. The only reason that it doesn't also mess with the ortho is probably because there is enough area outside of the problem surface that the pics can still be registered. You might try flying a mission at right angles to you original one or/and increasing the overlap. But I suspect that surface is still going to give you problems. Something you'll need to keep an eye out for in any other mapping that you do. In general though, topo errors on the order of 4 in or less (remember, even though the reference topo is very good, it also will have some error) is pretty good.

jephoto · May 25, 2017

What are the choices for a reasonable mapping software- no online fees- that I can practice with? Using P3Pro's & Mac computers?
Pro's & Cons?

Richard R · May 25, 2017

Check out the maps made easy app. Fairly inexpensive purchase and you can then upload and have process small maps for free. Limit is around 100 images. I've used them to map more than 20 acres at a time. Adjusting the altitude and overlap, I've gotten less then 2 inch pixels in the resulting ortho and kept under the limit. Even larger or higher resolution maps are pretty cheap. You purchase points and use them as needed.

PeteM · May 26, 2017

I tried to process the same set of images with Pix4D.

Aesthetically the result is worse, but at the level of measurements errors are better uniformly distributed.

In the area where Photoscan is 1 meter wrong with Pix4D we are between 0.2 and 0.5 m error. The overall metric accuracy is much better without distortions.

Richard R · May 26, 2017

Do you really mean 0.2-0.5 cm? I doubt that you can measure that finely on any topo map and it certainly wouldn't have that accuracy. Again, check out maps made easy. You can create an account and process images without using their app.The only issue would be that they recommend taking an image before the bird takes off to fly the mission. They then use that and the height info stored in the exif files to 'calibrate' flight altitudes. You can still create good maps, the reported MSL altitudes just won't be as accurate.

PeteM · May 26, 2017

Sorry... it was meter

With mapmadeeasy (free) I could export the dense point cloud in ply format ?

Richard R · May 26, 2017

PeteM said:
Sorry... it was meter

With mapmadeeasy (free) I could export the dense point cloud in ply format ?

No ply, but .las, .xyz, dem and .obj files

r62ewa · May 26, 2017

When I processed a job through Maps Made Easy and exported the DEM I believe it was actually a DSM.

Richard R · May 26, 2017

r62ewa said:
When I processed a job through Maps Made Easy and exported the DEM I believe it was actually a DSM.

It is. Real DEM takes some pretty sophisticated algorithms and processing.

r62ewa · May 26, 2017

Richard R said:
It is. Real DEM takes some pretty sophisticated algorithms and processing.

I know Pix4D has this function now but when they stated DEM on Maps Made Easy I was excited thinking they provided some type of automatic DEM creation like Pix4D but of course it was just the DSM. Would be great for those without experience to know it is actually just the DSM and not a DEM.

Richard R · May 26, 2017

r62ewa said:
I know Pix4D has this function now but when they stated DEM on Maps Made Easy I was excited thinking they provided some type of automatic DEM creation like Pix4D but of course it was just the DSM. Would be great for those without experience to know it is actually just the DSM and not a DEM.

Are you sure about Pix4D doing DEMs? All of the pages on their website refer to DSMs.

PeteM · May 27, 2017

Maybe I've found what's the problem.

Despite of great overlap (80% x 80%) in some portion of the white area the points that the software links together are not so much.

I can't see this problem on Photoscan but in the report of Pix4D there's a specific graph...

r62ewa · May 27, 2017

Richard R said:
Are you sure about Pix4D doing DEMs? All of the pages on their website refer to DSMs.

Yes, it was a feature they added not too long ago. I have watched a promo video of it. I do not have the licensed version to test unfortunately.

Richard R · May 27, 2017

r62ewa said:
Yes, it was a feature they added not too long ago. I have watched a promo video of it. I do not have the licensed version to test unfortunately.

Interesting. I'm not likely to ever be in the position to justify getting that version, so I wouldn't be testing it either. I can't imagine that they can do true DEM without a lot of user interaction. Someone/something has to classify all the points (e.g. Foliage,buildings, ground, etc.) before the algorithms that I am familiar with can develop the DEM. That portion of the process is, itself, pretty tedious. When I was still working Lidar for the USAF, we had a lot of pretty smart people looking at the problem. They were getting good results, but it wasn't easy or quick. If Pix4D really can produce true DEMs from your photos with minimum user interaction, I hope that they are marketing their services to all of those 3 letter government agencies. I knot that if I was still working at the AF lab, I would be very interested in talking to them.

r62ewa · May 27, 2017

Richard R said:
Are you sure about Pix4D doing DEMs? All of the pages on their website refer to DSMs.

Yes, it was a feature they added not too long ago. I have watched a promo video of it. I do not have the licensed version to test unfortunately.

Richard R said:
Interesting. I'm not likely to ever be in the position to justify getting that version, so I wouldn't be testing it either. I can't imagine that they can do true DEM without a lot of user interaction. Someone/something has to classify all the points (e.g. Foliage,buildings, ground, etc.) before the algorithms that I am familiar with can develop the DEM. That portion of the process is, itself, pretty tedious. When I was still working Lidar for the USAF, we had a lot of pretty smart people looking at the problem. They were getting good results, but it wasn't easy or quick. If Pix4D really can produce true DEMs from your photos with minimum user interaction, I hope that they are marketing their services to all of those 3 letter government agencies. I knot that if I was still working at the AF lab, I would be very interested in talking to them.

My mistake, Pix4D has an automatic DTM generation, you can view the process here DTM Extraction of Pix4Dmapper - Algorithm from Recent Research

PeteM · May 28, 2017

I did an experiment.

1) I made the point cloud, not georeferenced, not tied to the reality.
2) I imposed 5 points of known coordinates (4 corners + middle).
3) I applied an isotropic transformation to the entire point cloud.

Amazing precision

doktorinjh · May 30, 2017

Richard R said:
Are you sure about Pix4D doing DEMs? All of the pages on their website refer to DSMs.

Although there is some variation outside of the US, most photogrammetrists will agree that all DSMs and DTMs are DEMs. Digital Elevation Model is a generic term for Digital Surface Models and Digital Terrain Models. A DSM includes buildings and vegetation, a DTM is considered more of a "bare earth" type model. The USGS provides additional information (page 20) here. Pix4D seems to follow this nomenclature.