Constrained GCP's do not match Vertically in Generated DTM for Several CM's

WebODM
#1

I was performing an evaluation of a test ortho that I flew yesterday. The specifics of the project are GCP’s (existing catch basins) surveyed with RTK, Phantom 4 Pro flown at 200 ft. I generated the ortho and the DTM and then decided to check the elevations of the actual constrained control points to the DTM from inside of QGIS and noticed that although they should have been near zero in difference they were actually several cm difference as you can see in each of the below photographs. Additionally, horizontally they were off by 1-2 cm’s. Any idea why the vertical is not being constrained properly to the provided ground control? The red diamonds in the overview image are the 5 equally distributed GCP’s.

Overview of Constrained Points
Overview of Constrained Points1463×1535 744 KB

Constrained GCP 5 Off by 13 CM in DTM
Constrained GCP 5 Off by 13 CM in DTM2912×1692 183 KB
Constrained GCP 9 Off by 9 CM in DTM
Constrained GCP 9 Off by 9 CM in DTM2981×1583 190 KB
Constrained GCP 13 Off by 12 CM in DTM
Constrained GCP 13 Off by 12 CM in DTM2976×1605 170 KB
Constrained GCP 15 Off by 7 CM in DTM
Constrained GCP 15 Off by 7 CM in DTM2909×1692 170 KB
Constrained GCP 21 Off by 9 CM in DTM
Constrained GCP 21 Off by 9 CM in DTM2913×1607 209 KB

GCP File:

+proj=utm +zone=18 +ellps=WGS84 +datum=WGS84 +units=m +no_defs
587680.21 4732303.48 96.617 2524.85 246.15 DJI_0017.JPG 21
587680.21 4732303.48 96.617 2594.56 974.33 DJI_0018.JPG 21
587680.21 4732303.48 96.617 2688.35 1691.65 DJI_0019.JPG 21
587680.21 4732303.48 96.617 2820.00 2403.95 DJI_0020.JPG 21
587680.21 4732303.48 96.617 2965.26 3083.08 DJI_0021.JPG 21
587680.21 4732303.48 96.617 205.65 362.32 DJI_0041.JPG 21
587680.21 4732303.48 96.617 245.90 1078.90 DJI_0042.JPG 21
587680.21 4732303.48 96.617 268.15 1760.55 DJI_0043.JPG 21
587680.21 4732303.48 96.617 262.20 2472.60 DJI_0044.JPG 21
587771.34 4732470.93 96.383 871.09 891.36 DJI_0064.JPG 13
587771.34 4732470.93 96.383 812.80 1688.40 DJI_0065.JPG 13
587771.34 4732470.93 96.383 794.90 2467.20 DJI_0066.JPG 13
587771.34 4732470.93 96.383 813.15 3250.35 DJI_0067.JPG 13
587771.34 4732470.93 96.383 2156.40 1787.70 DJI_0117.JPG 13
587771.34 4732470.93 96.383 2034.95 2500.45 DJI_0118.JPG 13
587771.34 4732470.93 96.383 1909.40 3210.70 DJI_0119.JPG 13
587771.34 4732470.93 96.383 2373.90 365.05 DJI_0115.JPG 13
587771.34 4732470.93 96.383 2270.95 1076.85 DJI_0116.JPG 13
587830.93 4732328.84 95.723 896.90 421.60 DJI_0135.JPG 15
587830.93 4732328.84 95.723 980.25 1167.10 DJI_0136.JPG 15
587830.93 4732328.84 95.723 1080.70 1907.15 DJI_0137.JPG 15
587830.93 4732328.84 95.723 1205.30 2646.65 DJI_0138.JPG 15
587830.93 4732328.84 95.723 1347.55 3361.00 DJI_0139.JPG 15
587830.93 4732328.84 95.723 1964.65 332.60 DJI_0163.JPG 15
587830.93 4732328.84 95.723 1948.50 1048.85 DJI_0164.JPG 15
587830.93 4732328.84 95.723 1905.40 1761.30 DJI_0165.JPG 15
587830.93 4732328.84 95.723 1847.10 2480.15 DJI_0166.JPG 15
587830.93 4732328.84 95.723 1770.25 3204.60 DJI_0167.JPG 15
587833.60 4732204.73 95.417 591.60 333.40 DJI_0145.JPG 5
587833.60 4732204.73 95.417 765.05 1031.35 DJI_0146.JPG 5
587833.60 4732204.73 95.417 944.55 1760.65 DJI_0147.JPG 5
587833.60 4732204.73 95.417 1137.70 2473.65 DJI_0148.JPG 5
587833.60 4732204.73 95.417 1341.60 3168.70 DJI_0149.JPG 5
587833.60 4732204.73 95.417 1355.80 285.65 DJI_0152.JPG 5
587833.60 4732204.73 95.417 1716.40 769.15 DJI_0153.JPG 5
587833.60 4732204.73 95.417 1899.55 1477.65 DJI_0154.JPG 5
587833.60 4732204.73 95.417 2017.05 2172.60 DJI_0155.JPG 5
587833.60 4732204.73 95.417 2088.15 2870.50 DJI_0156.JPG 5
587937.43 4732401.31 95.486 4737.30 1391.05 DJI_0230.JPG 9
587937.43 4732401.31 95.486 4641.80 2112.80 DJI_0231.JPG 9
587937.43 4732401.31 95.486 4528.10 2868.40 DJI_0232.JPG 9
587937.43 4732401.31 95.486 2088.95 792.92 DJI_0250.JPG 9
587937.43 4732401.31 95.486 2101.85 1530.70 DJI_0251.JPG 9
587937.43 4732401.31 95.486 2141.60 2263.55 DJI_0252.JPG 9
587937.43 4732401.31 95.486 2212.10 2990.40 DJI_0253.JPG 9
587937.43 4732401.31 95.486 1032.82 831.33 DJI_0290.JPG 9
587937.43 4732401.31 95.486 923.30 1544.35 DJI_0291.JPG 9
587937.43 4732401.31 95.486 809.25 2232.45 DJI_0292.JPG 9
587937.43 4732401.31 95.486 687.90 2912.80 DJI_0293.JPG 9

3 Likes
#2

A few centimetres is pretty good. What are you trying to achieve?

The RTK coordinates are not a true measurement and if calculated over a UTM projection will also have scale factor differences at each different control point.

So survey grade RTK units are generally plus/minus 10mm for each antennae horizontally. Plus/minus 17mm vertically for each antennae.

Therefore at best RTK survey practice, plus/minus 20mm H and 35mm V at 68% confidence. Thats not even allowing for survey method, adjustment or analysis. Thats not allowing for difference to EXIF data rom the DJI JPEG.

I gather when ODM converges the images together it’ll find a ‘best fit’ datum and then calculate the vertical normals.

Better practice would be to use a survey Total station to measure your control horizontally,perform an adjustment and then perform a survey level run and run ODM without using the EXIF JPEG data which is assigned to image. Have all your survey control on the outside of the dataset, not within.

Always verify the data with survey method.

5 Likes
#3

Hi @AgingUETiger

That’s a nice orthophoto!

I’m not an expert but please give me the opportunity to share some ideas with you.

I have tested results processing same dataset and GCP on different software (Pix4D, Metashape and WebODM).
Processed the project 3 times with each software, and then compared the resulting 9 orthophotos.
Orthophotos showed slightly different results each time it was processed. Maybe I’m wrong but regarding the relative accuracy of the model you should expect the horizontal accuracy within the range of 2 times the GSD and vertical accuracy ranging from 3 to 4 times GSD, depending on the RTK equipment and procedures used. Remember that the absolute accuracy of the model can not be higher than the GCP’s accuracy, and for that reason it is important to measure GCP ensuring an accuracy higher than the GSD.

It is my opinion that the required accuracy of the survey job depends on the final application and use. Some jobs require sub-meter accuracy and some others will demand accuracy in the range of milimeters. So there are specific tools for each job.

Regarding the use of RTK, to increase accuracy of surveyed GCP then you can use static meassurements instead. Otherwise, like @Fletcher said, “Better practice would be to use a survey Total station”, but one must consider that the use of a total station adds costs and time to the field work. Please note that for some survey grade GNSS receivers, static meassurement accuracy is almost as good as using a total station (probably sub-centimeter accuracy). Its also important to note that often the accuracy is expresed as RMS, so for a GNSS receiver rated accuracy of 3mm+0.1ppm you should consider a 7.3mm+0.1ppm R95. But this is accurate enough for a 1 cm GSD.

I’m not really sure, but there is a possiblity that setting “texturing-nadir-weight” to 28 or 32, may help improve orthophoto accuracy for this particular case. I hope someone can clarify this.

Regards

2 Likes