Radiometric calibration options in version 1.4

[sean]

Hi

We currently use AgiSoft 1.3 and perform radiometric calibration ourselves using the documentation that Parrot and Micasense have released before feeding the images from the RedEdge and Sequoia cameras into Agisoft. We are interested in using Agisoft 1.4 to perform the radiometric calibration instead but we haven’t found enough detailed documentation so far in terms of exactly what calibration is carried out for both cameras.

From the AgiSoft  forum before 1.4 was released:

“We are working on radiometric calibration implementation in PhotoScan for MicaSense RedEdge and Parrot Sequoia cameras. Also there would be a possibility to use the irradiance information form the image meta-data.

Radiometric calibration of MicaSense RedEdge / Parrot Sequoia cameras would be supported in the version 1.4.”

Details in the user manual and release notes for 1.4 are very vague with no real details:

-   Added support for radiometric normalization based on camera exposure parameters and irradiance sensor meta data.

More forum info:

“The meta-information related to the irradiance sensor data should be taken into consideration during the orthomosaic generation process, if "Normalize Band Sensitivity" option is checked on in the Tools Menu -> Camera Calibration dialog -> Bands tab (for each band). The option is automatically checked on if the Reflectance Calibration procedure is completed.”

And finally a document posted to the forum – ‘Tutorial – Multispectral image calibration and processing in PhotoScan 1.4’.

So in terms of the MicaSense RedEdge camera do you perform all the corrections and conversion from raw DNs to radiance as per Micasense’s tutorial?

https://micasense.github.io/imageprocessing/MicaSense%20Image%20Processing%20Tutorial%201.html

-   Black levels
-   Vignetting correction
-   Row gradient correction
-   Raw DNs to radiance

Do you support the Micasense DLS (Downwelling Light Sensor – sun sensor) given that they haven’t published their DLS tutorial yet?

And in terms of the Parrot Sequoia camera do you perform all the corrections and conversion from raw DNs to radiance as per Parrot’s documentation?

http://forum.developer.parrot.com/t/parrot-announcement-release-of-application-notes/5455

-   Vignetting correction
-   Fisheye correction
-   Raw DNs to radiance

Do you support the Sequoia DLS? They don’t have an application note similar to the application notes above on their DLS but they have given out most of the required DLS information over a number of forum posts on their forum.

Since there is a ‘Use sun sensor’ checkbox in the ‘Calibrate Reflectance’ dialog I assume you support at least one of the manufacturers DLS?

When the ‘Use sun sensor’ checkbox is checked do you also take into account the sun’s azimuth and elevation at the time of the photos?

Lastly we’d like to confirm the options available with the various combinations of:

-   Normalize Band Sensitivity
-   Use reflectance panels
-   Use sun sensor

If we only select ‘Normalize Band Sensitivity’ I presume then that the corrections listed above, e.g. Vignetting correction, black levels etc. plus the conversion of raw DNs to radiance values will take place?

If we also then select ‘Use sun sensor’, but we don’t select ‘Use reflectance panels’ then the sunshine sensor’s data will be used in addition to/with the conversion of raw DNs to radiance values? Double-checking that the DLS (sun sensor) will/can be used even if ‘Use reflectance panels’ isn’t selected?

Thanks

[Alexey Pasumansky]

Hello sean,

PhotoScan 1.4.0 includes support for MicaSense RedEdge according to MicaSense recommendations (except DLS sensor support). PhotoScan 1.4.1 update will add DLS sensor support for MicaSense RedEdge camera.

Support for Parrot Sequoia in PhotoScan 1.4.0 is a little bit outdated. It will be updated to reflect current documentation from Parrot in PhotoScan 1.4.1 update as well.
Compensation by sun's azimuth and elevation will be performed starting from PhotoScan 1.4.1.

Reflectance calibration options in PhotoScan 1.4.1 for MicaSense RedEdge and Parrot Sequoia include:
- Reflectance panels only,
- Reflectance panels + DLS/Sunshine sensor,
- DLS/Sunshine sensor only.
"Normalize band sensitivity" option should be enabled in all cases. Selection between calibration modes is made using "Use sun sensor" and "Use reflectance panels" options from Calibrate Reflectance dialog.

In PhotoScan 1.4.0 the calibration options are limited to:
- Reflectance panels only for MicaSense RedEdge (conforms to current MicaSense recommendations),
- Reflectance panels + Sunshine sensor for Parrot Sequoia (not completely up to date).

[sean]

Hi Alexey

Thanks for the detailed response. Do you have an estimated date for when Photoscan 1.4.1 will be available?

And lastly in terms of updating your support for the Sequoia I presume you'll provide a mechanism to calculate the calibration between the Sequoia's multispectral sensors and the Sequoia's DLS, i.e. the K in the following formula from the following post?

K = Rref sunshine / sequoia

http://forum.developer.parrot.com/t/reflectance-estimation/5597/2

Cheers

[Alexey Pasumansky]

Hello sean,

I think version 1.4.1 will be released next week.

According to the provided link it is suggested to use the reflectance panel calibration:

One way of going around that is by using a calibration image with a target of known reflectance Rref, often a spectralon panel.

So it would be Reflectance panels + Sun sensor mode.

[glarsen]

Hi Sean and Alexey,

The same as Sean we have been applying corrections to our Micasense RedEdge photos as prescribed by Micasense.  One issue we are having is when importing the photos into Photoscan (v1.3.5) after corrections are applied it does not recognize the  images as multispectral images, although we have copied the entire exif header (using "exiv2 -ea" to extract header from originals and "exiv2 -ia" to insert in corrected) and kept the names of the corrected images the same. 
So what we have been doing is loading each band as a chunk, and then aligning chunks and I find it messy and not as accurate.  What are we doing wrong here Alexey?

Regards,
Glen
 

[Alexey Pasumansky]

Hello Glen,

You can send five images (from each RedEdge camera band) to support@agisoft.com, so that we can reproduce the issue on our side and find our the source of the problem.

[sean]

Hi Alexey

According to the provided link it is suggested to use the reflectance panel calibration:

One way of going around that is by using a calibration image with a target of known reflectance Rref, often a spectralon panel.

So it would be Reflectance panels + Sun sensor mode.

However the difference is that you only need to do perform the calculation of K (for a given Sequoia multispectral and DLS combination) once or possibly periodically like every 6 months or so as opposed to for every flight.

So the idea would be to cache the value of K for a given Sequoia unit and then use it when processing data from this unit for multiple flights without a reflectance panel.

So I was wondering whether PhotoScan 1.4.1 would have the ability to calculate K and then store/cache it for a specific Sequoia unit to use it when processing data from that specific Sequoia unit from further flights that don't include reflectance panel images?

If you look further down the thread that I linked to you'll see me asking Parrot about the frequency of calculating K and Parrot's response:

Doing it for each flight is likely overkill.
Indeed, many aspects are difficult to control in the field and render a target measurement counterproductive:
Casting shadows or reflecting light onto the target (note the equations above have sun identical for both sensors for a given acquisition, a bright red shirt can screw up the measurement).
Not considering angle of observation (can be really bad when cos(theta) plays a role).
Having substantially less than the entire hemisphere for Sunshine or target (if one stands quite close to the drone, then the geometries are not comparable. This is more dramatic under purely diffuse light)
On the other hand, the only real way of ensuring that your data is consistent through time is measuring a known standard. This applies generally to any precision measurement device.
The frequency of calibration (measurement against a standard) will depend on the precision that you require.

Cheers

[glarsen]

Hello Glen,

You can send five images (from each RedEdge camera band) to support@agisoft.com, so that we can reproduce the issue on our side and find our the source of the problem.

HI Alexey,
I just sent 10 images.  5 are original files recognized by Photoscan as multispectral images, and 5 processed images not recognized as multispectral images.
Regards,
Glen

[SvaneSDU]

I would sign up for an implementation of DN to reflectance conversion directly in Photoscan. I've tried the Micasense tutorial but streamlining the workflow through photoscan would be great.

[Leo7best]

Hello guys.

I use version 1.4.1 PS

I have the Parrot Sequoia Camera and I'm trying to calibrate the images created by Ebee SenseFly Drone.

During calibration and creation of the clipping mask, the system does not create the  Panel Calibration.

How can I solve the problem?

Attached a screen "Cattura.png"