Field Camera Calibration Procedure

[Darko]

Hi,

I just want to know that I am doing it properly and if you have any suggestions you are welcome.
I am using PPK GNSS positions for my cameras so I've used the following procedure.

1. All the camera coordinates were imported together with the x/y/h precision.
2. Camera calibration was loaded from my previously adjusted job done with the GCP's.
3. Camera was set to PRECALIBRATED with FIX calibration checked.
4. Cameras were aligned with HIGH precision and GENERIC preselection.
5. 60 field measured GPC's were imported with the precision of +/- 1cm evenly spread all over the calibration field.
6. GCP's were moved to it's real position and overall precision for the control was about 0.30 m.
7. FIX calibration tag was removed from the camera calibration.
8. OPTIMIZE cameras was started with all the corrections checked.
9. Total error on control points was now 0.07m (25mm pixel size).
10. In CAMERA calibration I've switched to Adjusted values and saved the calibration in a file.

The same area was photographed once again just 10 min. after the first flight. Calibration that was saved in the previous procedure was used to redo the job. Now I've stopped on the 6th step and my precision showed 0.10 m.

Is this procedure OK? If not, where I've made a mistake and what's the way to get even better precision without GCP's?

Best regards,
Darko
 

[JMR]

Questions: How do you transform antenna coordinates to camera center coordinates? is your camera attached to the drone frame rigidly (no gimbal)?
which camera/lens are you using?
and with regard to the lens pre calibration... how was the flight designed and how does look like the calibration field? if you trust the precalibration... why do you release calibration in step7?

[ruyi7952]

I am using PPK and have achieved satisfactory accuracy. My method is the default workflow. =)

Don't be confused by "accuracy" at any Settings dialog box..

A high-precision control point is more helpful than a camera calibration.

[Darko]

Hi,

Antenna is attached above the camera gimbal (+23cm). Actually it is behind it for 9cm but when flying at the speed of 12m/sec it is above it. Could be that it is fluctuating few mils but it is not that important.

Camera is Sony A6000 with the 20mm prime. Camera is set to MANUAL with the aperture 8 and exposition 1/1250. That's fixed, the only thing that I am changing is ISO sensitivity.

Calibration field is 250 m wide by 500 m long. The flight speed is constant, set up to 12m/sec. My drone is X Hexa multirotor with the regular servo gimbal with roll and pitch compensation. All the GCP's were measured with the RTK GNSS instrument and result is the mean of 10 measurements.

In step 7 I've taken off FIX calibration in order to calculate new calibration parameters that were saved in the file. In the next process where I've checked my camera calibration, those NEW calibration parameters has been taken as FIXED and I have just taken my original GCP's as a CHECK points to see how far are my results from the original GCP coordinates.

The mean error was better than I've written it before, it was +/-0.062 m. In that check process I have used another set of photos taken few minutes after the first one (second flight).   

The goal that I want to achieve is to have a good precision model without the need for any GCP's. Actually I always have one and that's my take off point (white square table with the black square in the center). Since my camera is always started before flight I always have few coordinates for that point that is 32 cm below my GNSS antenna.

I've tried to work without fixed calibration and in that case my control was worse than 1m, compared to 6cm that I have now.

[ruyi7952]

My experience is that the high accuracy of control points and the high quality photos, is more important than the camera calibration parameters, because the software will have free-net camera calibration。

Please don't worry about camera calibration.

If your shutter speed is 1/1250.

1、Change the flight speed of 12m/s to 8m/s.
2、The aperture is adjusted to 5.6.

[Darko]

Hi Ruyi7952

The reason I am flying with 12m/sec is the area coverage. Very often my coverage is more than 2000 ha per day. The aperture 8 is the best setup for this 20mm Sony lens, 5.6 is also OK but loosing sharpness in the corners. My regular flight is 270 m above the ground with the ground resolution of 5 cm.

When shooting large forested areas there is no chance to make any GCP and that is the reason that I want to have reliable model without GCP's.

[Hydro_Ydé]

Hi Darko,

I do not work with PPK yet, but I tried also to determine the best camera calibration parameters. What I did was to disable all camera location (uncheck camears on reference panel) and keep only GCP locations, then run alignment. I found that I get much more constant camera parameters like this.
Working with P4P, I have sometime very large error in Z cam, and even with large values given to camera accuracy, final solution is greatly affected by these bad locations.
And of course it is very important to have a good set of camera parameters, in particular F value, who lead to large Z error if badly determined.

[JMR]

Hi Darco
I'd recommend you to try the voightlander Heliar 15mm which is a manual lens that can be turned into true fixed lens easily. On the contrary, the sony 20mm otherwise is convenient due to its small size and and lightnes is not very sharp and, more importantly, it is hardly fixable.
May I ask what ppk system are you using? we are producing our own one, it boasts a dual frequency receiver, and as far as I know it is the only one existing that uses two inertial units one on main board and the other on the camera hotshoe to monitor both the drone and camera attitudes independently, thus making posible a true per-shot offset correction.
kind regards

[Darko]

Hi JMR,

I am using North Survey RTKITE GNSS RTK module. Actually I am not using RTK since it is too complicated so I am using PPK mode which is much simpler for the field use. There is no IMU installed on it. You don't need it at all since the PhotoScan takes those photo centers as the base and with the fixed camera parameters it just finds the best rotation angles by itself. Without any GCP's I've come to precision of 6 cm which is perfectly OK for my use. If there is a need for better precision it is good to have few GCP's. What I noticed is that now I have much, much better elevation precision.

My GNSS antenna (2 Freq., GPS + GLONASS, 400 channels), is located 23 cm above the camera center (center of gimbal rotation) and 9 cm back and when it flies with the speed of 12 m/s it is right above the camera.

I've tested so many camera lens configurations and I can say that the best result was with Samsung NX500 + 20-50 mm plastic lens that came with cheaper Samsung. Unfortunately Samsung is not fireing sync signal with the shutter speed higher than 1/250 and I had to discard it.

Combination that I have now is the lightest version that works with RTK. I've bought Sony A7 R2 with 42 Mpix resolution and seems like I am going to buy Zeiss Loxia 21 mm for it. The only problem is that I have to construct and print the new bigger and stronger gimbal and it's gonna be at least 600 gr heavier. So my flights will be shorter but pictures will be bigger. I guess that will be my final investment in this drone.

Now I can fly for 40 min before forced landing. All my plans never exceed 20 min of real flight so I am always on the safe side.

RTKITE is very cheap but working solution (2000 Euro). The company is located in EU, Barcelona. Juan that is the technical support guy is very kind and wants to help.

There is a cheaper solution but it is from China and the company is TERSUS and model is PRECIS BX316R. This GNSS has two antenna connectors for heading determination. I've bought one, but didn't have time to do anything with it. It looks nice and promising and when I get a chance to test it I can get back to you. Do you want to buy it from me? I'll sell it for the price I've paid for it. I think it was about 1000 Euro.

With the RTKITE I had a problem that lead me to full drone reconstruction. It didn't want to work with LK900 Data link that I had with my DJI A2 controller. After replacement of DJI with Pixhawk and with even better and stronger data link it works fine.
 

[JMR]

"You don't need it at all since the PhotoScan takes those photo centers as the base and with the fixed camera parameters it just finds the best rotation angles by itself."
Well, Darko, you might be wrong here. Photoscan can handle and calibrate fixed offset on its own, notice that in the camera calibration dialog there is a tab for offset, but it can't when camera mount is not fixed in the same rigid body as the antenna.
That's why we use two inertial units in our solution Drobit. Is not for the calculation of camera angles as it seems you have understood. As you correctly say, yaw pitch roll are calculated by Photoscan during orientation, but our goal it is to translate coordinates from the antenna phase center to the camera's nodal point, which generally is not that simple as to subtract 23cm from Z (or more precisely the hypotenuse 24,69 according to your numbers).
by monitoring frame and camera attitudes for each shot, we can calculate exactly xyz components of the offset (vector camera-antenna) for any camera pose regardless the flight speed or drone attitude even with the gimbal having three degrees of freedom and rotates eccentrically, which is a general case.
Drobit has been recently chosen by Airbus, and is also used by some other companies that have understood the importance of the feature that I tried to explain. By the way they are getting up to <3cm accuracy

[Darko]

Hi JMR,

You are absolutely right about getting good camera offset parameters but, what is the precision of my GCP's ? I am measuring it with RTK GNSS that is held by hand and I can say it is no better than +/- 3-5 cm accuracy. Who can hold it absolutely still? Than, what is the absolute precision of my drone GNSS? It shows something like +/- 1 cm, but do you really believe in that? What is my time offset between the taken picture and recorded position ? Not sure about that, but I am flying with the speed of 12 m/s which means that in 1/1000 sec my path is 12 mm? My gimbal is not precisely below my antenna but that fluctuation is better than the real precision of the GNSS.

So, to make it simpler I've just decided to subtract those 23 cm (hypotenuse) from the antenna elevation to get assumed projection center coordinates and to check what can I get with it. I am sure that someone can get those 3 cm but I am perfectly happy with 6. Who is measuring points on the ground with better precision? I know, sometimes it must be done but in that case I'll do it by my surveying total station. 

Any other solution is much more complicated and it must have IMU's or the other type of drone and gimbal construction, so you can say it is some kind of trade off to get it as simple as possible but loosing some precision.