Mavic 3T Thermal Processing

[rhprecisionunmanned]

Hello Community!

So we have recently acquired a Mavic 3T drone for conducting thermal scanning. We have had great success with processing Index Maps in Pix4D with other platforms (Intel Falcon 8+ with Flir Tau2, M300 with FLIR Zenmuse Xt2 and H20T when images were converted). I know that Pix4D does not support the M3T for radiometric processing as the files are encrypted by DJI. That being said I have found a 3rd party converter that will create FLIR radiometric image files that can be viewed in FLIR tools. The converted data sets are compatible with Pix4D and the ‘FLIR extraction’ does occur once the images are uploaded. Everything works very well up until the point where the software reads the EXIF data from each image - I believe the way the images are coded by the converter negates some of the information Pix requires for the best outputs. I am able to process all the way through to index map generation however the software creates a completely flat surface with no elevation differences. This is fine for doing ag work but it created all kinds of issues and distortions for roof inspections where there was a 20m difference between ground/roof.

I have used a company for converting H20T images where the integration is seamless (no elevation/point issues) however they do not offer the same service for M3T data. This is the error message I am receiving once processing has started.

[Warning]: Warning w9020: GDAL Warning <1> TIFFFetchNormalTag:ASCII value for tag “ImageDescription” contains null byte in value; value incorrectly truncated during reading due to implementation limitations

[Warning]: Warning w9020: GDAL Warning <1> TIFFFetchNormalTag:ASCII value for tag “Make” contains null byte in value; value incorrectly truncated during reading due to implementation limitations

At first I wondered if adding the camera model would help or using a similar template (I think i tried a Flir 640 with same lens size) however I got the same messages for each image. Does anyone know if there is a workaround for this or if there is a step in pre processing that can negate the Exif Data? I unfortunately had to jump to Drone Deploy to meet my timeline requirements for the project as the M3T (without completing a conversion) is supported on that platform for processing. Hoping someone at Pix4D can shed some light as I find Pix4D is a far superior product overall and do not want to have to shift to DD for processing.

Thanks for the help!

[tom1]

Do you absolutely need the radiometric data for the orthomosaic? I recently was experimenting with creating an orthomosaic of a large apartment complex we did IR roof scans on with our M3t. The software had some issues when processing as a thermal project. After some quick research I ran the IR images as a regular RGB project and the resulting Orho was acceptable for our needs.

[rhprecisionunmanned]

Hi Tom,

Thanks for the reply. Unfortunately we have been generating categorized Index maps for the last 5 years and our clients have come to expect the outputs with temperature readings. I have done some comparison samples in the past and got much different results between a processed RGB map vs a radiometric index map from a visual perspective. We found that with an Index map we had a ~90% accuracy for thermal anomaly identification whereas it was closer to 40-50% with strictly a visual RGB map. The problem we find is that depending on the flat field correction of the sensor there can be major variances between images as the sensor picks up new hot or cold features between images. Even with 90% overlap we had areas that the RGB map would convey what appears to be an issue but when compared with the radiometric map it was simply washout over larger area from one specific location (ie vent, AC unit, HVAC etc.).

Cheers and thanks for the input

[Kapil_Khanal]

Hello, Can you explain what issue you are having with PIX4Dmapper? Are you having an issue with the quality of the result or having issue of the software crashing? Usually, a warning message shouldn’t cause the project to fail.
With thermal images, what output are you generating and what are you looking for as an output?

[rhprecisionunmanned]

Hello Kapil,

Thanks for reaching out. I put up a detailed post on the community board as you suggested and reached out to Daniele. We create index maps for our clients under different thermal palettes and categories. The software is not failing - just providing sub par index maps with gaps/distortions. I know that it is not a ‘Pix’ issue per say but rather the way the images are writing the EXIF information upon conversion to RJPEG for Pix compatibility. What I am really hoping to get from you and Pix is some guidance on how I may be able to work around this issue - we were forced to switch to Drone Deploy as a processing solution as we were on a tight deadline and did not know how long it would take to find a solution - the data sets processed and created a radiometric Ortho from the original M3T (non-converted) images. We are much happier with Pix4D as a product and have been using/promoting the software for several years now. I am hoping there is something that I can do in terms of adding a step to my worklfow/upload to circumvent the issues I am seeing with the final outputs.

Thanks for the assistance

---

Edit: Added private message (Daniele)

The product that we use to convert is called ThermoConverter - great product for creating RJPEG files that are compatible with FLIR Tools. After conversion I go through the typical workflow for creating a project under the thermal camera template. However when I get to the Image Properties screen I notice that the GPS coordinates are incorrect (not a Pix issue but the converter). My way around this is creating a CSV file in EXCEL with the original image data and importing the Geolocation and Orientation data direct from a log file. This works for correcting the X,Y information. Once I get to the main screen I begin Initial Processing and immediately receive the error message posted above. It appears that the camera model information and TIFF tag have eliminated some information that I believe is critical to processing - at least from the outputs I am getting. Essentially Pix is ‘flattening’ the entire surface without any elevation changes (the sample data is a roof about 20m above grade). Pix is able to extract the FLIR data and knows to define the images as greyscale and gives the following.

The software does not fail - it continues through tie point reconstruction as well as point cloud generation. Here is what I am seeing in terms of a ‘flattened’ surface.

Ultimately we provide 3 different categorized Index maps to our clients in different colour palettes (usually 5, 15 and 32 categories to show macro vs micro anomolies) but the 3rd stage of processing is having issues and I am assuming that it is having trouble building the reflectance map as it is declaring the surface as a single elevation.

I was on a short timeline for this particular project and unfortunately was not able to submit a support ticket/deal with the issue at the time. We acquired a Drone Deploy license as it did support the processing of M3T thermal data sets for creating radiometric maps. The data sets processed well however we are more familiar and happier with Pix4D outputs. I am hoping that there might be a way to ‘trick’ the software or work around the error message so that we can move back to Pix4D as our processing solution. Here is the quality report - I will need to post the Log separately as it is too large of an upload.

Quality Report.pdf (516.0 KB)

Log Report.zip (1.05 MB)

Thanks, and have a great day.

[Kapil_Khanal]

Hello @rhprecisionunmanned,

1. Regarding the point cloud being flat. This might be the issue due to the camera optimization error as seen in the quality report.
   
   

To resolve the issue, use the All Prior setting as shown in the figure below.

2. Regarding the distortion/gaps, it is mainly due to the quality of the images. I would recommend using 90% image overlap and a flight height of 50 meters or higher. If you are mapping the roof of 20m then fly at 70 meters or higher.

3. The reflectance map is not the same as the orthomosaic. They won’t have evlevation data on it. You should be using a point cloud to visualize it in the 3D.

As a conclusion, I don’t see a big issue with your project. The distortion is due to the quality of the images, regarding the point cloud being flat, I belive using All Prior should resolve the issue if there is enough image overlap as required.

[keith2]

I’m a relatively new user of mapper, could you please explain the process of being able to visualize the index information as flat or in 3d in software that is shareable with clients? I have exported a geotiff but I’m not quite sure which of the 11 files outputted are necessary to view this is a shared web based environment.

[Mike_K]

Hi Keith,
You might want to look at PIX4Dcloud. This will allow you to share your results in a web-based environment. The article below describes how you would upload the results. Please not that this does require an additional license to use the service.

How to upload or replace results to PIX4Dcloud