I've heard this line repeated time and time again. I'm currently testing the accuracy of RTK only flights vs RTK + GCPs vs GCPs are varying density.?ÿ
I'm curious, does anyone know where this comes from and the logic behind it?
I'll be releasing my results in the next week from what I'm seeing, in the meantime I was hoping to hear other peoples observations.
Haven’t done RTK but tested PPK, short story…still need GCPs. PPK might screw it down to a foot or two but that’s about it.
In my experience with over a thousand flights verified in the past six years I can say it is true. Today I often use just 5 targets on most flights. I will mark them as check points, and find the horizontal to be in tolerance 98%, and the vertical is in tolerance 75%. When the vertical is out of tolerance this is most often a constant. Which can be adjusted without reprocessing, using Global Mapper or TopoDot. Pix4dmatic has finally included the Geoid18 and it works very well. I always fly with RTK, PPK, and a few GCP's for redundancy. If your tolerances are failing with RTK/PPK you likely flew too high for the camera, with poor GSD, or camera over lap is spaced too far apart, or poor lighting or wind or other factors that can be mitigated with experience.
Haven’t done RTK but tested PPK, short story…still need GCPs. PPK might screw it down to a foot or two but that’s about it.
Any chance you have time for the long story?
In my experience with over a thousand flights verified in the past six years I can say it is true. Today I often use just 5 targets on most flights. I will mark them as check points, and find the horizontal to be in tolerance 98%, and the vertical is in tolerance 75%. When the vertical is out of tolerance this is most often a constant. Which can be adjusted without reprocessing, using Global Mapper or TopoDot. Pix4dmatic has finally included the Geoid18 and it works very well. I always fly with RTK, PPK, and a few GCP's for redundancy. If your tolerances are failing with RTK/PPK you likely flew too high for the camera, with poor GSD, or camera over lap is spaced too far apart, or poor lighting or wind or other factors that can be mitigated with experience.
What aircraft were you flying and what are your tolerances?
So far I have done 3 separate flights, 3 different days and 3 different areas. One was 100 acres, the other two were about 20.
The smaller two I had 5 check points, the larger I had 21.
Im seeing an average delta hz dist of about 20-30mm, same with vertical (slightly worse). Did I just get extremely lucky? Have UAV's changed that allow for better processing?
I even turned off the base station on one flight for the last 15% of the images. The check points that were covered by these were out by about 60mm but still not too bad considering that were using the base station corrections at the time I turned it off (several minutes later).
I flew with a M3E and processed with Metashape.
What aircraft were you flying and what are your tolerance
Most of my flights were with P4RTK with Pix4dMapper. Last few months I'm using M3E with Pix4dmatic. My tolerance is 0.10ft (30mm)
I find that Pix4D highly over-exaggerates the horizontal error when reporting the RMS value. In the attached screenshot the N & E mean are 0.012 feet (4mm), with the Z mean of 0.057ft, with an overall mean checkpoint RMS error of 0.037ft. However, if you see large horiz mean values (like your 20-30mm) then something is wrong. Either with the flight, the rtk, ppk, or your coordinates for the checkpoints.
The Pix4dmatic report is very poor, compared to Pix4dMapper.
What aircraft were you flying and what are your tolerance
Most of my flights were with P4RTK with Pix4dMapper. Last few months I'm using M3E with Pix4dmatic. My tolerance is 0.10ft
I find that Pix4D highly over-exaggerates the horizontal error when reporting the RMS value. In the attached screenshot the N & E mean are 0.012 feet, with the Z mean of 0.057ft, with an overall mean check point RMS error of 0.037ft. However, if you see large horiz mean values (like your 20-30mm) then something is wrong. Either with the flight, the rtk, ppk, or your coordinates for the checkpoints.
The Pix4dmatic report is very poor, compared to Pix4dMapper.
I'm a bit confused, you mention "if you see large horiz mean values (like your 20-30mm)", you consider these large? They fit within your tolerance of 0.10ft.
Yea I wouldn't trust getting my "errors" from the photogrammetry software. I could definitely see them over exaggerating the quality.
To get my delta hz I picked out the check shots in the final ortho in C3D and created a point. To get my delta verticals I picked the nearest point in the point cloud and used the elevation of that in the picked point in C3D and made some tables in excel comparing those coordinates to the averaged GPS rover shots. This method I feel directly compares the output data (what we actually care about) to our best indication of where that point should be (I was seeing about 7-15mm in between RTK rover shots from different epochs)
Bit of an offside, do you prefer Pix4dmatic to the pilot 2 app? I've been extremely happy with the pilot 2 app.
I'm a bit confused, you mention "if you see large horiz mean values (like your 20-30mm)", you consider these large? They fit within your tolerance of 0.10ft.
Bit of an offside, do you prefer Pix4dmatic to the pilot 2 app? I've been extremely happy with the pilot 2 app.
To me, the horizontal error in Pix4D seems to be exaggerated by 5x. Therefore if your Pix4d residuals are 30mm, then your error could be something like 150mm. Obviously, my RTK is not within 4mm, and we certainly can't mark photos at that precision either.
You may be confusing these software applications. Pix4dmatic is a PC program for Post-Processing drone images, which will someday replace Pix4dMapper. The Pilot 2 app is Android and used to fly the drone mission, yes this is what I use. There are third-party drone apps such as hammer, drone deploy, and UGcS but I don't trust them, as they are still in the development stage for the M3E. Let someone beta-test that, not me.
My residuals aren’t 30mm. My actual accuracy is that (worst check points)
With my WingtraOne PPK I rarely need GCPs. Most of the time I put out targets only for checkpoints. Horizontal rms is 1x gsd and vertical is 1-1.5x gsd on bare ground with good texture. Perfect for civil earthworks construction.
Haven’t done RTK but tested PPK, short story…still need GCPs. PPK might screw it down to a foot or two but that’s about it.
Any chance you have time for the long story?
We are using a Freefly Astro with Sony Alpha 7R (61mp). We are using the Freefly PPK tool. It could be their implementation is less than ideal. Other manufacturers may produce better results. Other than that the product we are getting is excellent. Much better than the dji Phantoms which we are often prohibited from flying.
Attached is my report. I did further checking of the target in the orthomosaic and using the point cloud for elevation and the residuals gathered that way very closely match Pix4Dmapper's reported errors.
Haven’t done RTK but tested PPK, short story…still need GCPs. PPK might screw it down to a foot or two but that’s about it.
Any chance you have time for the long story?
We are using a Freefly Astro with Sony Alpha 7R (61mp). We are using the Freefly PPK tool. It could be their implementation is less than ideal. Other manufacturers may produce better results. Other than that the product we are getting is excellent. Much better than the dji Phantoms which we are often prohibited from flying.
Attached is my report. I did further checking of the target in the orthomosaic and using the point cloud for elevation and the residuals gathered that way very closely match Pix4Dmapper's reported errors.
Pix4DMapper doesn’t support local geoids. That’s probably why your vertical errors are so large. I use TBC to convert the geotags from lat long ellipsoid height to state plane NAVD88 elevation. Then import that csv into Pix4DMapper. Input and output coordinates are the same (State plane) then pick arbitrary for vertical coordinates. They way Pix4D doesn’t perform any horizontal or vertical projection. This should provide much better results.
Pix4DMatic does support geoids, so the workflow is more streamlined than Mapper.
I've heard this line repeated time and time again. I'm currently testing the accuracy of RTK only flights vs RTK + GCPs vs GCPs are varying density.
I'm curious, does anyone know where this comes from and the logic behind it?
I'll be releasing my results in the next week from what I'm seeing, in the meantime I was hoping to hear other peoples observations.
We are now flying RTK as a first option, more convenient & efficient in term of data storage management etc. PPK is the next option.
We no longer use GCP per say, but more of as checkpoints for verifications.
No longer trust GCPs only methodology. The result tends to 'drift off' if the GCPs are not well distributed.
I've heard this line repeated time and time again. I'm currently testing the accuracy of RTK only flights vs RTK + GCPs vs GCPs are varying density.
I'm curious, does anyone know where this comes from and the logic behind it?
I'll be releasing my results in the next week from what I'm seeing, in the meantime I was hoping to hear other peoples observations.
We are now flying RTK as a first option, more convenient & efficient in term of data storage management etc. PPK is the next option.
We no longer use GCP per say, but more of as checkpoints for verifications.No longer trust GCPs only methodology. The result tends to 'drift off' if the GCPs are not well distributed.
Absolutely, I actually made a video comparing accuracy a few months back after I made this post...
@bc-surveyor
Excellent work and thank you for sharing.
That is not even considering the time and effort to establish, not only enough number of GCPs but also to ensure that they are well distributed over the entire project area. We discovered this during the old Phantom 4 days a few years back, unable to gain access to one of the corners of the project area to survey 1 GCP in, but proceeded to process away... and ended up with busted heights of almost 0.5m on that particular corner..scary stuff , luckily it was discovered early and only costed us another 200km trip back to site.