Please select Into the mobile phone version | Continue to access the computer ver.
P4M Camera shutter issue - Am I the only one?
1926 19 2020-12-15
Uploading and Loding Picture ...(0/1)
o(^-^)o
Digisky
lvl.2
Flight distance : 91411 ft
New Zealand
Offline

Hello,

I recently got my P4 multispectral and did 8 flight tests so far.
The aircraft flies well but I have an issue with the captured imagery. When looking at the photos, I noticed that, randomly, some photos (only with monochromatic sensors) were highly underexposed.

I did several tests with AUTO mode for the NDVI sensors and with shutter priority (at 1/400) but the problem remains.
Even when the shutter speed is set at 1/400, some photos are being captured with a shutter speed over 1/2000 (see screen capture below).

No need to say those underexposed images make the processing and the final outputs highly inaccurate and therefore unusable.
Am I the only one with this issue?



Shutter Priority set at 1/400

Shutter Priority set at 1/400
2020-12-15
Use props
ds9193
lvl.2
Flight distance : 64469 ft
Australia
Offline

Hi Digisky,

It looks like the automatic metering in the camera may have adjusted the shutter speed for the bright shiny line in the middle of the image. Notice once it passes the centre of the image, the exposure returns to normal? This is because of the centre weighted averaging, which calculates exposure based on the average pixel values in the centre of the frame.  An exposure lock here would be useful, though I don't think the Multispectral sensor has that ability..


Cheers,

Dan

2020-12-15
Use props
Digisky
lvl.2
Flight distance : 91411 ft
New Zealand
Offline

Hi Dan,

Thank you for your reply.
Your explanation makes sense if you work on AUTO mode. In my particular case, the camera is set to shutter speed priority, therefore, the camera should stick to the speed I set.
This is why I think there is something wrong with my camera.
I am still waiting for a reply from my supplier.

Cheers

Moe
2020-12-16
Use props
Geebax
Captain
Australia
Offline

Digisky Posted at 12-16 19:33
Hi Dan,

Thank you for your reply.

This would not be the first time a DJI camera implemented an un-wanted auto exposure function. Try testing it indoors to see if it alters the exposure automatically when confronted with a high brightness object.

It is highly unlikel;y that your supplier will know anything about thye subject, you are far more likely to get useful advice from this forum.

2020-12-17
Use props
DrMrdalj
Second Officer
Flight distance : 1265978 ft
  • >>>
Serbia
Offline

Digisky Posted at 12-16 19:33
Hi Dan,

Thank you for your reply.

Please note that shutter priority is still auto-mode, while it favors variation of the aperture. If aperture is constant (fixed in lens or already maxed) then shutter will still get shorter if brightness input gets stronger.
What makes me more puzzled is the impression that NIR and RE channels did not get same exposure change as it was the case with B, G, R channels. can you confirm if I am right?
Having said that, there should be all the details of the exposure recorder in XML/EXIF, so software can compute actual reflectance and therefore proper vegetation indexes. Have you processed these images in DJI Terra? In not so, have you performed reflectance calibration of the images prior to processing in some other software like Metashape?
2020-12-18
Use props
Digisky
lvl.2
Flight distance : 91411 ft
New Zealand
Offline

Geebax Posted at 12-17 13:22
This would not be the first time a DJI camera implemented an un-wanted auto exposure function. Try testing it indoors to see if it alters the exposure automatically when confronted with a high brightness object.

It is highly unlikel;y that your supplier will know anything about thye subject, you are far more likely to get useful advice from this forum.

Hi Geebax,

Thanks for the reply and advice.
I've done it your way by putting the aircraft on a table and facing a hedge.
I then triggered a photo every 2sec with shutter priority set to 1/400.
All works fine until I walk slowly in front of the camera holding a white towel.
I am ending up with underexposed photos (for the towel pictures only) and when looking at the EXIF data, the shutter speed didn't remain at 1/400, it jumped to the roof.

I agree with you that the DJI supplier is not the greatest support, but at least they are in touch with DJI and they will provide them hopefully feedbacks.
2020-12-21
Use props
Digisky
lvl.2
Flight distance : 91411 ft
New Zealand
Offline

DrMrdalj Posted at 12-18 07:01
Please note that shutter priority is still auto-mode, while it favors variation of the aperture. If aperture is constant (fixed in lens or already maxed) then shutter will still get shorter if brightness input gets stronger.
What makes me more puzzled is the impression that NIR and RE channels did not get same exposure change as it was the case with B, G, R channels. can you confirm if I am right?
Having said that, there should be all the details of the exposure recorder in XML/EXIF, so software can compute actual reflectance and therefore proper vegetation indexes. Have you processed these images in DJI Terra? In not so, have you performed reflectance calibration of the images prior to processing in some other software like Metashape?

Hi DrMrdalj,

Thanks for your post.
I agree with you about shutter priority being still an auto-mode but for every camera that I had in the past if I set a shutter speed priority, let's say at 1/400, every photo I'll take will be at the same shutter speed.
But this camera won't.
Regarding the NIR and RE channels, you are right, they don't seem to be affected. They appear to be much more sensitive as well (their shutter speed is way higher when sunny and  with Auto-mode).
See the attached image, with the shutter speed for each channel under sunny conditions. Interesting to find out that the Blue channel is way less sensitive than the NIR or RE channels.


I have processed the image with Terra and I notice some processing errors particularly visible with the OSAVI index (see below).
I've also processed the images with Pix4D fields and I got similar results.
I don't use Metashape as I've learnt that the software doesn't read the EXIF data from the P4MP.

Anyway, I cannot supply any outputs in these conditions, there are too many errors due to the underexposed images.
2020-12-21
Use props
diegogw
lvl.1

Peru
Offline

Hi there. I'm having the same problem. Did you manage to solve it?

Best.
2021-5-16
Use props
Digisky
lvl.2
Flight distance : 91411 ft
New Zealand
Offline

Hi Diego,

Unfortunately, no. DJI won't acknowledge the issue and they keep saying this is normal.
I even traveled to the DJI dealer to compare my drone with their multispectral drone. They both behaved the same and the issue happened as soon as a bright object crossed the frame. The guy I was dealing with didn't understand anything about photography which is crazy since they are the main resellers in NZ and they basically mostly sell flying cameras.
Bottom line, I stick with my drone and I shoot in full auto mode (as advised by another user). I am being very carreful with the data I get around a bright object, which is not very often fortunately.
Let me know if you manage to get better info.
Cheers
2021-5-16
Use props
LV_Forestry
First Officer
Flight distance : 4726654 ft
Latvia
Offline

Hi,

I use auto mode for the shutter, and also auto white balance. The shutter speed changes very often during the flight. I sometimes have exposure differences, but not as much as yours. On your NDVI raster it is obvious that there is a problem. I am using Metashape, it works great. If you wish you can send me your photo set via a google drive, I will put it in Metashape we will see the result.

Raster NDVI

Raster NDVI



Jordane



2021-5-16
Use props
LV_Forestry
First Officer
Flight distance : 4726654 ft
Latvia
Offline

I managed to reproduce your problem by parking my vehicle in the middle of the field to be scanned, which I usually never do. SurexpNDVI.PNG SurexpMS.PNG


It clearly creates a big problem on the calculated indexes.
I tried on another field, same results.
Surexp 2.PNG

What I see is that the presence of clear color (My car is clear grey) metallic/plastic materials promotes this effect. The concrete does not seem to have an effect.


2021-5-18
Use props
Digisky
lvl.2
Flight distance : 91411 ft
New Zealand
Offline

Hi Jordane,

Thank you for your message and for being thorough in your investigation.
Well, I am happy I’m not the only one to think there is an issue with the camera.
The issue happens whenever a bright object (even small) crosses the frame. In my case, this is a big problem since all the orchards in my area have sheds or houses nearby (with clear roofs) and vehicles could be found between blocks.

Anyway, no need to fly to replicate the camera issue: I left the aircraft on a table, facing a green hedge, I triggered the camera every 2sec. and I slowly walked pass the camera while holding a white sheet of paper (actually, wearing a white T-shirt works as well). As soon as I appear in the frame, the shutter speed will jump and underexpose my image.
Interestingly this only affects the Red, Green and Blue sensors.

What I find particularly annoying is the camera won’t do what it is supposed to do.
I tried to lock the shutter speed with “shutter speed priority”, but it won’t…
I also tried the “AE lock” feature but nothing is locked and the shutter speed keeps playing around.
And DJI finds this 100% normal, and my local reseller doesn’t even understand what I mean.
This is particularly frustrating and disappointing when you consider that you are paying a premium price since it’s “DJI Entreprise”. I was hoping a better support but there’s none; that’s the DJI way.
And I don’t even mention how frustrating it is to have to fly with DJI GSpro; I hate this app and that’s the only one we are able to use.

That is interesting to see that you have good results with Metashape. I didn’t know it could deal with the DJI exif data.
I also notice you use QGIS, is there any interesting tool that could be used with NDVI imagery? What file format are you using to import in QGIS?

Thanks

Moe
2021-5-18
Use props
LV_Forestry
First Officer
Flight distance : 4726654 ft
Latvia
Offline

Hi Moe,

Before acquiring the phantom 4 MS I was working with a Tetracam ADC. I got into the habit of inserting the sorting step into my workflow.
I was hoping that it would be different with the P4 but unfortunately not!
I can only recommend that you complete this step and delete the overexposed images.

About DJI, I imagine that there is at least one person who fully understands all this.
At least the one or those who designated the drone and his camera, that we can only greet them because we must admit that a multispectral camera of this quality in a "ready to fly" drone for this price, with this autonomy in flight, RTK ready, it's a great succes!
Unfortunately these people cannot be contacted directly, we often have to deal with people who are on the whole very knowledgeable about "consumer" products. But for dedicated applications like multispectral, there is still work to be done. If DJI donated $ 1 to charity every time they wrote the word "sorry", we would be close to finding solutions to many problems in this world. Failure to respond to our problems regarding their drones. Hey DJI, that's humor, please don't make my drone aggressive on me after the next update !
About the local dealers, personally I don't think we can blame them. they sell what DJI gives them to sell and they certainly don't have time to test everything.

I find that overall GSPro does the job, in comparaison with the first GS working with 2.4 Datalink. I just find it sad to pay for a 6000euros drone and to have to add 50euros to be able to upload KMLs and shapefiles in this app.
There is an alternative on android. It is not yet perfectly developed, I discovered it recently and am still in the testing phase. This is UGCS. Even though this is a company also from Latvia, I have no connection with them, they haven't given me anything to say that, but I find the support very responsive and the software very comprehensive. There is a full demo version that runs for 15 days if you dare to try it. It does not solve the problem of overexposure, but it is a professional alternative to GSPRO for P4M. Just need a laptop and android phone / tablet.

About Metashape, the same thing, they gave me nothing to say what follow, we chose this software because it was the cheapest at the time. I have been using it for about 2 years for 2D / 3D / Multispectral reconstruction applications of forests and agricultural fields. I am fully satisfied with it, and my colleagues too. For having also tried the PIX4D software suite, I can not find anything that makes me think of giving up Metashape. Metashape seems to integrate perfectly the EXIF data of DJI as well as the correction of the reflectance given by the solar sensor integrated in the drone. I say "seems" because multispectral data is always subject to interpretation given many influencing factors, the position of the sun, the weather ... But at first glance our data are consistent. What I find good in Metashape is the possibility of including the geoid of your region, this makes the geographic correction, notably on Z, excellent. There was no such possibility in the version of PIX4D that I tried, at least I couldn't find it. Let's not even think about it in Terra.

About DJI Terra is good if you want to quickly see the state of your crops. Possibly detect an anomaly, but it doesn't have the power of Metashape and QGIS.

Speaking of QGIS, this is the final product that allows us to analyze, compare, in detail the rasters in geoTIFF format that are extracted from Metashape. It is possible to perform a Raster transformation immediately in Metashape in an unlimited way, because you can manually enter known formulas as well as your own formulas depending on the application. But you can also do it in QGIS.

There are nevertheless little tips to know to use your rasters optimally, you can find on the internet many scientific publications on this subject. My job is to operate the drone and do the pre-processing of the images until providing rasters. Afterwards I have colleagues who are very proficient in analysis. Do not hesitate to contact me if you have any other question, I can try to help you find the answer.

Regards

Jordane



2021-5-18
Use props
Digisky
lvl.2
Flight distance : 91411 ft
New Zealand
Offline

Thank you Jordane for taking the time to provide so much info.

I agree with you on many points.
-        I think DJI should stick to Hardware manufacturing as I never been satisfied with their flying Apps. Even when we had to use the Datalink (with Wookong M), the open source software such as Mission Planner (Ardupilot) was and still is much better. Moreover, DJI was charging over $1000 for 20 waypoints.
I discovered GSpro 5 years ago when I purchased my first P4pro and I quickly swapped to Litchy.
When I got the P4MS, I was very disappointed to find out that Litchy wouldn’t work with this aircraft and that GSpro hadn’t improve at all in the meantime.
I have 4 main issues with GSpro:
1.        I cannot work from my desktop
2.        I cannot change the elevation of an individual waypoint: I’m based in NZ, mostly doing orchard mapping (kiwifruits and avocado trees) and it’s very common to find important height variation within the same orchard.
3.        I cannot fine adjust each waypoint to ensure I am not flying over the neighbour’s orchard.
4.        The orchards that I survey are far from being a square shape, and the grid that the app. will generate tends to give a weird flight path (with crossing lines).
With Litchy, I have to do my flight grid by hand but I have full control of each waypoint and the app even uses Google earth elevation to follow the terrain. Litchy works from a web page on my desktop and the flight plan I save can be found on my Ipad, very convenient.
I might give a try with UGCS if that works with the P4M. I am surprised though since I thought DJI had never supplied a SDK for their Entreprise range.

-        I never tried Metashape but I had the opportunity to compare its older version (Photoscan) a few years ago with Pix4D and I had better result with Agisoft than Pix4D. I found it more user-friendly as well (NDVI was not involved at the time).

I have a monthly subscription with Pix4Dfields so I’m not stuck with them. Still trying to figure out which one (between Agisoft and Pix4D) will suit my needs best. DJI terra is not on my list, this sofware is a joke.

-        I see your company seems to have quite some experience in that field and I was wondering whether you were using a reflectance calibration target for each of your flights?
Do you need to have the same weather (sunny or cloudy) when you do consecutive survey over the same area? I am very confused on how consistency can be achieved over time and there is no true answer I could find.

I sincerely appreciate your help and will eventually PM you (since this thread is becoming off topic) if you don’t mind.

Kind regards

Moe
2021-5-19
Use props
LV_Forestry
First Officer
Flight distance : 4726654 ft
Latvia
Offline

Hi Moe,

You are welcome, I work for a research institute, my goal is not to amass a maximum of money, if I can store knowledge by restoring that which was given to me, I am happy .
Regarding your issue with GSPro, If you want to read below I will explain how I do it. It's not a universal truth but it works for me and it might give you some ideas.
50% of my mission preparation takes place on QGIS.
-Taking into account the cadastral plot or part of it to be studied
-Taking into account the surrounding elements, Housing, power lines, prohibited surface ...
-Drafting of authorization requests, NOTAMs, information for residents with the mapping tool.
-Cutting of the cadastral vector then import into GSPro (additional function which costs 50 euros)50% remaining in GSpro is set the overlap, the altitude regarding the cm/pixel requested, safety check, flight !

Regarding the 3-dimensional flight plan, UGCS does it without problem. You can even import a DSM raster from a previous photogrammetry to increase the precision. Have you tried the 3D Map Area function of GSpro? Latvia is a relatively flat country, not being faced with this problem I have no solid experience on it.

Metashape is in my opinion now much better than Photoscan. Fortunately because it is the evolution of this software.

Terra I imagine it well for the independent farmer who owns his own drone and does not want to spend time playing with rasters. He clicks on NDVI, green is good red is not good. This may be more than sufficient in certain applications.

About the reflectance calibration panel, to be honest I don't use it every time since I have the solar sensor on the phantom, although they are two different things anyway.
If you want to get similar results between two consecutive flights, you will need to use a reflectance panel to calibrate the sensitivity of the camera. The solar collector is more there to compensate for variations in exposure during the flight.
But again, what is the goal?
If for a very specific reason you want to get the perfect index yes it is better to use a panel. But are we sure of this perfect index?
Take the case of the most famous NDVI that gives you the stressful state of your plant.
Hypotesis: 0 she is done / 1 she is in great shape.
Finding: Healthy plants generally return indices around 0.8

NDVI DIFF

NDVI DIFF


I put you an example in photo, the left part of the forest has on average an index of 0.8, you perceive very clearly that the closer you get to the river on the right side the more the index reduced. it drops to 0.4.

On the DSM we can see that the trees have the same volume, grow in the same way as the others.
So lack of water? Too much water? Fungal attack? Insect?
No ! different species of tree which had not finished budding at that time.

This is why, I personally find that the tools of PIX4D and Terra for the classification of zones by index are good, but we quickly made amalgamations. Nothing beats a well-cut raster well documented on QGIS to know what you are looking at. You understand that if the index is 4 or 4.5 it doesn't matter. Especially since we must not lose sight of the fact that the plants move with natural phenomena such as the sun, the wind, throughout the flight. So in a laboratory reproduce an index on a leaf whose exposure does not change why not. But in our case I don't see the point!
As for the weather you know here in Latvia from October to June we don't see the sun often. It's not a problem. Once again if you are organized you will know how to treat your data as it should.
Shading can sometimes be a problem when I need super quality DSMs. I just do the flight within half an hour after sunset if it's a day when we don't have a cloud. I reassure you that does not happen often!


You can send me examples of your orchards, I will try to help you optimize your flight plans.

Everything that has been written is a purely personal opinion and adapted to our activity and our environment. It is certainly not a universal law. I invite those who read to react in their turn, to give us their point of view.

Have a nice day !

Jordane


NDVI DIFF.PNG
2021-5-20
Use props
LV_Forestry
First Officer
Flight distance : 4726654 ft
Latvia
Offline

In addition :
If I had to redo the field you gave as an example, I will prepare the flight in this way: fpl.PNG
Then then in QGIS I will do a vectorization in this way to separate the plots and adjust the calculation :
QGIS vect.PNG
If you removed your overexposed images at begining of the process, you should be able to obtain a satisfactory result despite the overexposure effect of the sensor. But from what I can see on satellite images it has a sort of white bar that reflects all of the light and acts as a target for recovery.
Then you can go further by making a grid and asking QGIS to give you the average of the indexes in a radius around each point. This is just one example! it all depends on what you want to achieve.

grid.PNG

Hope this helps, but in summary there is no quick fix to remove this overexposure effect. For now it requires a little work in post production. But I'm sure DJI will do something ...

Jordane

2021-5-21
Use props
Digisky
lvl.2
Flight distance : 91411 ft
New Zealand
Offline

Hi Jordane,

Thank you again for all these valuable info.
I’ve uploaded the images for this orchard. It’ll only comprise the top of the orchard, since the compressed file is already over 3Gb. You will notice that the shutter speed issue happens quite often. The white bars you mentioned are artificial shelters, to protect the vines from the wind; they are very common in kiwi fruit orchards.
This flight was done with AUTO settings. I flew over the same orchard with different settings (shutter speed priority, AE lock) and the result was always the same. It’s like having adjustment settings available but they don’t affect the camera at all, it keeps shooting like it is in AUTO mode no matter what.

I will PM you the link to upload the file since I am not too comfortable sharing it on the web.

I am not super happy to remove any image as in my case that will represent a huge amount of them and it will obviously affect the desired overlap.
I could still adjust with photoshop the exposure for each underexposed image but that would be such a tedious task; and not efficient at all.

As you said, DJI will hopefully do something but so far they didn’t even acknowledge the issue and to be honest, I have the feeling we are only two to agree there is something wrong.

Thanks again for the help.

Moe
2021-5-23
Use props
louishouette
lvl.1

France
Offline

Hi guys,

Thanks for a very interesting talk. I'm having the same problems. It seems so far that the only way for me to get a correct result (regarding the exposure) is to use Open Drone Map.

I'll keep watching this thread.
Louis
2021-7-12
Use props
djiuser_t7hibdA1RlE8
lvl.1

United States
Offline

Glad I found this thread. I also have been having the same issues with flying the P4M and having some images over or underexposed as the camera moves through a scene.

From the bit I know about photography this clearly is an issue with the metering algorithm on the sensors. It seems to me that being able to change the metering from center-weighted to averaged/evaluative would go a long way to mitigate this issue. It's disappointing to me therefore that different metering options aren't available in the GS Pro app when almost every digital camera you can buy these days has that setting. Maybe metering for multispectral sensors is more complicated and that's why we don't have the option?

Really hoping DJI can release a patch that addresses this. I wouldn't say that it has outright ruined any of my surveys but it certainly has diminished the results in certain areas.
2021-8-3
Use props
patiam
Captain
Flight distance : 1037995 ft
  • >>>
United States
Offline

This thread is AWESOME! Thanks to @Digisky (Moe) for starting it and reporting the issue, and especially to @LV_Forestry (Jordane) for all the very helpful info! (Jordane you've given me advice in other P4MS threads). Lots of very important and useful info here... Keep it coming!

I particularly like (and agree) w/ Jordane'spost #13 on 5/18 regarding the realities of DJI's dev and support.

BTW Pix4D does allow correction from HAE to Geoid but only with a single value, as opposed to allowing the use of a raster to apply a correction that varies over space. But for most UAS mapping projects in most locations, the Geoid height does not vary much over the area mapped.

(I still haven't flown our P4MS yet, waiting on an iPad so I can fly it w/ it's "preferred" app before trying UgCS per Jordane's suggestion. We're a primarily Android shop, so I'd rather not use an iPad unless it is demonstrably better.)
2021-8-6
Use props
Advanced
You need to log in before you can reply Login | Register now

Credit Rules