I am trying to get adjusted and get my workflows together using the RTK. I have a flown a project before without the RTK enabled and the result was as eexpected. Yesterday I flew a new project and decided to use the Mobile Station with the RTK fix.... Brought the data back into the office and ran the Pix4D on it....
This is what I get? I assume that the Mobile station is worthless and I should just use the PPK post processing....

The Blue dots are the actual image locations. The Green ones are the adjusted ones. What makes them act that way?



Looks like Bending data..... Around a imaginary globe....

Hi Ishka,

The blue dots, being the "actual image locations" suggest that your problem is not with data collection (ie drone and base station), but with processing in Pix4D. It very much looks like there is something seriously off with your Step 1. Initial Processing settings.

What settings have you got under the Calibration tab for step 1.?

Regards,
Roger

dodger_rog Posted at 8-20 15:21
Hi Ishka,

The blue dots, being the "actual image locations" suggest that your problem is not with data collection (ie drone and base station), but with processing in Pix4D. It very much looks like there is something seriously off with your Step 1. Initial Processing settings.

Well, the whole setup is messed up. I mean the images from any dataset that involves the use of the DJI Mobile Station in RTK mode with the FIX had this issue.... It is not a matter of the software.... I tested the same dataset in ESRI Drone2Map, same issue.... I then took another stab at another site and flew it without RTK and everything looks fine.... I will test the same site again later with RTK mode again, but not messing with any "aides" in processing, such as the toggles that enhance your imagery, distortion and so.... Little by little.... But everyone seems to favor the PPK processing versus the RTK with mobile station, so pretty much I should not even bother with the tripod.... But if they say that RTK mode works too, I need to figure out under which conditions it does...

And here I was thinking that you were asking for help.

As an FYI, we routinely fly with the RTK Base Station over the same site approximately every 2 weeks and get sub 2cm errors every single time.

Well, any help would be welcome... Just trying to figure out if I should leave any settings off when capturing data? I am trying to figure out if the DJI Mobile Station can be used.... Or should I skip using it and just apply PPK correction afterwards.... Because as I see out of the box, using the right workflow, when using the RTK the data is messed up once the Calibration applies in both Pix4D and Drone2Map.... Without RTK the data gets the right output.... So maybe as you mentioned it is something in the software that readjusts the data and makes the green go nuts?

So the issue it's either in the field collection or the initial processing settings...

dodger_rog Posted at 8-23 16:35
And here I was thinking that you were asking for help.

As an FYI, we routinely fly with the RTK Base Station over the same site approximately every 2 weeks and get sub 2cm errors every single time.

Here are the settings under calibration... I left everything as default...

Ishka Posted at 8-23 17:07
[view_image]

Here are the settings under calibration... I left everything as default...

And I guess this is the default for the data? Do I need to select the corridor, if it is a mix of NADIR and obliques? Does it make a difference?

dodger_rog Posted at 8-23 16:35
And here I was thinking that you were asking for help.

As an FYI, we routinely fly with the RTK Base Station over the same site approximately every 2 weeks and get sub 2cm errors every single time.

So my workflow was the following:

Set the DJI Mobile Station and turned it on. All good on that.
Turned the Controller on.
Turned the drone on.
All linked together
RTK on and the controller shows FIX.

Mission fly and return back.
Downloaded the images and created a new 3D project in Pix4D.
Used the default settings for processing and the output for the first step had the issue....

So either the FIX did not work as expected in the field, either I am missing something in step 1.
Thanks for all the help... Drone2Map uses the same engine as Pix4D, so maybe that is why I had the same result in both....

I did another session without RTK and did not see this issue in processing using the default settings for the project. So maybe for the RTK session I need to define something custom in the step 1 for processing, or maybe something was off for that day with RTK....

Hi Ishka,

I see two settings in Pix4D that you can change first off.

1. Under the Matching tab, Matching Image Pairs - change this to Aerial Grid or Corridor;

2. Under the Calibration tab, Camera Optimisation -> Internal Parameters Optimisation - change this to AllPrior.

Try changing these first and see what output you get.

Further investigation may require looking into your specific drones camera values . The default camera settings in Pix4D for the P4RTK drone camera may be too far away from yours to produce accurate, reproducible results. If you are using the default values and want to change them, I can talk you through how to find out your Drone's calibrated values to input. A lot of people have had problems before they changed these values.

Regards,
Roger

Hey Roger,

While waiting for the answer I was playing around with some settings and I did the following:

I changed the Matching to Aerial Grid or Corridor
I changed the calibration from Standard to Accurate Geolocation (something like that). I cannot stop the process now and see the exact name.
Then under the Internal I changed the Setting to All Prior.
For External I left it as default.

While waiting for the second step I got the report at the first step and things seem to come together.









So things seem to look better on the first look. The report typically doesn't tell me too much since I am new to the Pix4D game, but I see good values for the errors... Or at least it looks to me. So I guess I did the same changes you mentioned, with just one extra change....

So now I just try to make sure that I have the right step by step for each situation:

1. For NADIR Shots I would use the aerial grid corridor with the modifications for the calibration.
2. For mixed NADIR and oblique, what should I use? Same configuration?

So was I correct to change the setting from Standard to the Accurate? Since the data should be accurate with the FIX RTK? Or should I leave it as standard and change only the parameters you mentioned?

Do I need to use the same settings whether RTK or PPK data? Does this modifications apply only for the case of HQ geolocated data that comes from P4RTK or should I apply them also for data from for example Inspire 2 Pro? I guess these settings seem to have to be modified for data that doesn't need optimization from the field? P4RTK data should be that way, while other data would not need these changes and just use the defaults...

How would I also know if my camera settings are correct, or should I modify them for processing? I want to be accurate, so if the default values from the camera database are off I would want to know what I should use, or where to find the ones I am supposed to use and how should I apply them. So would appreciate your guidance.

Thanks a lot for all the help.

With your workflow, the only things we do in addition to yours (and you may also be doing this, so apologies if you do and have left them off):

1. On jobs where we have known Ground Controls Points, we set the base station up on one of these and enter the x, y and z coordinates into the drone. Note: Coordinate input is in WGS84 in decimal degrees, height is height of the GCP plus height of the base station (1.802m).

2. Once everything is linked, we let the base station collect data for ~10mins before we launch.

As for your question regarding NADIR or obliques, I've processed both types together with the same settings as that which I use for NADIR only and get great results. So I don't think that is an issue.

dodger_rog Posted at 8-23 18:24
With your workflow, the only things we do in addition to yours (and you may also be doing this, so apologies if you do and have left them off):

1. On jobs where we have known Ground Controls Points, we set the base station up on one of these and enter the x, y and z coordinates into the drone. Note: Coordinate input is in WGS84 in decimal degrees, height is height of the GCP plus height of the base station (1.802m).

So should I leave the calibration to Standard or Accurate Geolocation and Orientation?

I am not using the known GCP yet, but I will be doing that in the future...For now I was just setting the base station and then left it for 15 minutes or so on before I was flying. I did not care for now about the dead on accuracy, but something that gets closer to that without GCP...

So the height of the GCP (Z value) is based on WGS84, I assume.

So do I need to change the values of the database camera, or should I stick with the default?

Leave the calibration setting as Standard.

Yes, the height of the GCP is the WGS84 ellipsoid value (this is what the drone uses).

For changing the values of the database camera, I would definitely recommend it if your aim is to eventually have accurate and reproducible results. However, let's see what changing the initial processing settings does first.

Hi Ishka,

I missed your last lot of replies, just reading through them now. The output looks heaps better!

With the Accurate Geolocation and Orientation setting, my understanding is that this assumes your image capture is absolutely accurate in terms of geolocation and orientation, so it forces Pix to change other things to get a good result. I would recommend leaving this set to Standard (even with accurate geolocation info, we leave it as Standard and get good results).

Your Sensor Dimensions output in the Quality Report is quite telling. Pix is changing the focal length significantly and is also changing the radial and tangential distortion values quite a lot. I would definitely recommend putting your calibrated values in the camera database.

Let me know if you want me to talk you through this.

Roger

dodger_rog Posted at 8-23 19:30
Hi Ishka,

I missed your last lot of replies, just reading through them now. The output looks heaps better!

Yes, if the camera data needs to be calibrated and updated to get accurate results, I would want to know how to do that. Is there a database online of what those values should actually be? or it is more of a research from trial and error?

Here are the results from a cross grid project, 60 degree shots with the same settings I had before...



I will run another session with the settings you provide and change it to standard and provide the output data then.

As a side question.... Do you use the same settings for other drones as well? If you fly a Phantom 3 or Inspire 2 Pro with an X5S, do you use updated values for the camera and same settings for calibration? or those apply only to Phantom 4 RTK?

Ishka Posted at 8-24 04:40
Yes, if the camera data needs to be calibrated and updated to get accurate results, I would want to know how to do that. Is there a database online of what those values should actually be? or it is more of a research from trial and error?

Here are the results from a cross grid project, 60 degree shots with the same settings I had before...

Hi ishka,

To get your camera settings, you'll first need to use an exif reader online. I find this one really good:

http://exif.regex.info/exif.cgi

Upload ine of your images to this, doesn't matter which one.

Look for the values under:
Calibrated Focal length
Calibrated x
Calibrated y; and
Dewarp data (has a date and then 9 values, the last 5 are what you want). The last 5 values are the R1, R2, T1, T2 and R3 in order.

When you load a Project in Pix4D, where it has the camera there should be an Edit... button on the righthand side. Click this and then click New under the camera type.

On the lefthand side (think this is the "image" rather than "pixels" enter the calibrated focal length, x and y in the boxes and then enter the R1 etc values in the boxes below.

Rename the camera and save. Set this camera as default.

I'm not sure about any other drone models, we only fly the P4RTK sorry.

Any issues, let me know. I'm on the road for a couple of days but will answer when I can.

R

dodger_rog Posted at 8-24 17:13
Hi ishka,

To get your camera settings, you'll first need to use an exif reader online. I find this one really good:

1. I will check on getting this set. The data I flew has a DEWARP setting of 1 and it doesn't show those values, since I enabled the distortion correction on the settings when I flew it. I however found that data on other images that were flown without the distortion correction enabled. So I will try to define that camera.

2. If you define that camera using the calibrated data, do you still use the Standard with All Prior? Because you don't want PIX4D to do too much optimization...

3. By the way, on the data that I had from the camera I tried to use the standard and it did not work well, so for this specific dataset, I had to use the last option of accurate geotagging and orientation. Maybe it was because of the dewarping embedded in the image?

4. Also, do you guys fly RTK or PPK for the data? I guess if I remember well, you fly RTK using a known point.

5.If you plan to fly an area and try to get the best model with data under the trees how do you plan that using DJI4RTK? For the overall I guess NADIR is all good. How do you plan to fly the oblique? I know I can fly multiple missions at different angles of the same area. If I fly without a plan, the data doesn't contain any GPS data. So what do you use for the flight on custom path, or maybe around the objects? The corridor path with manual camera? or is there any other mode to account for flying around the objects? Or I am stuck with the do your best from multiple angles with the cross grids?

Hi Ishka,

1. Looking at your results you posted I was wondering whether or not you had enabled the distortion correction when capturing the images as I had never seen those values go to zero before. I've never used the distortion correction. I'm not completely clear as to the why of it, but I was of the understanding that having the distortion helps with the 3D-ness of the model. I'm sure more learned ppl on this forum could explain it...

2. Definitely with the calibrated camera, keep using Standard and All Prior. Using Accurate Geolocation forces Pix to find other solutions when the best/correct solution maybe simply moving the "camera". All Prior forces Pix to use camera calibration values that are close to those that you have input.

3. Not too sure about this one, but you might be right about the distortion correction.

4. We always fly RTK, we don't do any PPK. It is worth mentioning however, if you don't set up the Base Station on a known location (and input the location into the remote controller) you will get a survey where each individual images location is very precise with respect to each other, but the absolute accuracy of the model could be out by up to 2-3m in x, y & z. This is where I suspect PPK is advantageous as you can get a correction in post processing. The advantages of RTK when you set up on a known point means that there is no PPK processing.

5. We don't have too much experience with big trees, we usually fly NADIR at a flight height of 100m. For the majority of our work, we fly using terrain awareness. Trees aren't an issue for us. I would suggest having a play with the different planning modes and see what you get. You could try flying at 2 different heights at 2 different camera angles. I can say though, that flying double grids (ie the 3D grid) you do get a better point cloud classification than what you do with the 2D grid.

Hope all this helps!

Roger

dodger_rog Posted at 8-25 15:00
Hi Ishka,

1. Looking at your results you posted I was wondering whether or not you had enabled the distortion correction when capturing the images as I had never seen those values go to zero before. I've never used the distortion correction. I'm not completely clear as to the why of it, but I was of the understanding that having the distortion helps with the 3D-ness of the model. I'm sure more learned ppl on this forum could explain it...

Preparing this week to do another set of flights and process data. I got my survey guys to get me some control points on site so that I can use the known control point data.

So my plan was to fly the site using different configuration and heights. The only issue I will have is how to fly the random areas, since in order to get more accurate data on the model I need to fly  oblique shots of the tree areas. The other issue I have is that if I fly in manual mode, there is no RTK mode on that mode, so how would I deal with those photos in relation to the RTK photos? Or all the dataset needs to be RTK, or can I use a combination of RTK photos and non RTK photo and let Pix4D take care of the issue?

I am planning to do the processing in Drone2Map, Pix4D, Metashaper and ContextCapture and see what is the overall product I could use for the future. last time I checked, Pix4D was the overall winner, with ContextCapture doing the best about the densification of the point clouds. I like that metashaper can define a KML plan for flight for an area so that you can get a more accurate model.

So in the field, I will use the known control point with the defined lat and longs and for height I will use the meter height of the point plus the height of the rod? Then leave it for 10 minutes to collect data before I make my first flight? How did you find the 1.802 for the height? it was not defined anywhere....

Ishka Posted at 9-14 06:08
Preparing this week to do another set of flights and process data. I got my survey guys to get me some control points on site so that I can use the known control point data.

So my plan was to fly the site using different configuration and heights. The only issue I will have is how to fly the random areas, since in order to get more accurate data on the model I need to fly  oblique shots of the tree areas. The other issue I have is that if I fly in manual mode, there is no RTK mode on that mode, so how would I deal with those photos in relation to the RTK photos? Or all the dataset needs to be RTK, or can I use a combination of RTK photos and non RTK photo and let Pix4D take care of the issue?

Hey Ishka,

I'm pretty sure you can still "free fly" in RTK mode. I've done free flying after surveys to drain batteries and it doesn't mind at all. I can't however be certain that when you are free flying the photos will be accurately geotagged in the exif metadata (I would just have a go and see what you get).

The Base Station Height been fairly widely discussed in numerous other posts on here. I can't remember if that is where I originally got the 1.802m value from but it works well in our work flow. Just remember that you want to add it to the Ellipsoid Height of the surveyed Ground Control Point you are setting up on, not the height in another datum.

Good luck, hope it goes well!

Roger

dodger_rog Posted at 9-14 19:15
Hey Ishka,

I'm pretty sure you can still "free fly" in RTK mode. I've done free flying after surveys to drain batteries and it doesn't mind at all. I can't however be certain that when you are free flying the photos will be accurately geotagged in the exif metadata (I would just have a go and see what you get).

How do you set it up over a known location? I don't seem to find a place where you can type in the known location. Or is this done afterwards in post processing?
Update: Maybe I think I have found the location under Advanced Settings in the RTK menu. So do you recommend setting up on a known control point? or set it anywhere and use the GCPs for tie at the end?

Definitely set it up over a known point and enter the coordinates. And it sounds like you have found where to enter the coordinates, under the RTK menu, go to advanced settings and input GNSS coordinates. Should be in decimal degrees for lat/long and metres for height. Then in the main RTK menu, check the coordinates are now correct.

If you don't set it up on a known point, you can bring it back to where it should be using the GCPs in processing. The problem with not setting up on a known point is that the project will be out (in absolute terms) by up to a couple of metres. Not such a problem when you have the base station set up on a single point continuously, but if you turn the base station off, or move it to another location, then each of these "sub-projects" will be differently offset by up to a couple of metres.

What we now do is setup on a known point, enter the coordinates and then after step 1 processing we will add the other GCPs as 3D control points and then re-optimise once all these have been entered. In the beginning, we didn't do this and had an extremely difficult time getting the "sub-projects" to align during processing. Several times we ended up having to go back out and re-fly the project. I can't tell you how much time I spent trying to align projects initially (think weeks/months!). I ended up figuring out a solution, but it is far easier to set the known location first!

I've just re-done one of these last week. Approx. 5,500 photos, drone launched from 4 different locations (ie 4 different known base station sites). But because I entered the coordinates for each of these during setup, the project stitched together seamlessly...

dodger_rog Posted at 8-24 17:13
Hi ishka,

To get your camera settings, you'll first need to use an exif reader online. I find this one really good:

So I am trying to define my camera and you mentioned about the Calibrated Focal length, X, Y and the 5 dewarp data numbers.

This is the camera as defined by the PIX4D


This is what I get from the EXIF reader


As you can see the Calibrated x and y values on this are thousands away from the small values they are defined by PIX4D.

Here is the expanded Dewarp Data



So what should I define based on this data for the optimized camera?

Ishka Posted at 9-21 06:11
So I am trying to define my camera and you mentioned about the Calibrated Focal length, X, Y and the 5 dewarp data numbers.

This is the camera as defined by the PIX4D

So I guess, I found you need to click on the image and define it as this?



Are my values accurately defined based on the previous post?

Ishka Posted at 9-21 06:24
So I guess, I found you need to click on the image and define it as this?

[view_image]

Hi Ishka,

Correct, looks like you got them all in the right spot!
Regards,
Roger

Getting to process the data in between the other projects I need to deal with.... I am testing three pieces of software to process the same dataset and see what do I get out of it.... (Pix4D, Metashape and ContextCapture). It seems that one most versatile and providing most customization of the data is Metashape... Will provide more info as I see things happen. It seems that I also have to deal with the difference between ellipsoid height and geoid because they do not take that in acccount. I guess I need to figure out the average height difference for the site and apply that to that processing to actually get some accurate data...