Alignment of low-res thermal images

[Sune]

Hi

We want to create thermal orthomosaics. We have data captured with a Tau2 336 sensor, but it seems the resolution is too low for proper alignment.

Because we are screening the market for the best suited software for this purpose, it would be great with some guidelines from the support team.

I have sent 2 small data sets to support@agisoft.com.

Can you please check if alignment is possible with the radiometric jpegs in the data sets?

Next week we will recieve our new tau2 640 sensor, but we still have a huge chunk of thermal data we would like to create thermal orthomosaics, from the 336 sensor.

Kind Regards
Sune

[Alexey Pasumansky]

Hello Sune,

We have received the data but also haven't achieved good alignment results with that. Probably, having side overlap would improve the alignment stability, do you have such sample sets, where there are at least two-three flight-lines with side overlap over 50%?

Another potential problem that you may encounter, even if the alignment is completed, is low number of tie-points for 0.1 MPix images - it will lead to the impossibility to generate the dense point cloud.

[Sune]

Hi Alexey

Thank you for the quick response. Here are a few more questions.

The data sent to you is extracted from a .SEQ file. We store 8 images/sec, but skip 3 and 5 frames in the data sent to you. Any chance more images will improve the result (larger frontal overlap)?

Why does PS need the side overlap to align the images? What other options are available for aligning data sets with no side overlap? A low-tech solution could be mounting 2 thermal sensors on the same gimbal, creating a data set with the desired side overlap. Could this solve the problem?

I will send you a sample set, captured with a 640 sensor. It covers a small area, 1 flight line. Frameskip i 4. PS successfully aligns the images without any reference file loaded. If i compared the 640 data with the 336 data i sent you, the only "real" difference, besides the geografical location, is the sensor resolution. Is the conclusion that a 336 sensor simply has too few pixels, for proper work/output in PS?

Is it possible to use another pallet on the thermal images, creating higher contrast in the images for the tie points/alignment? I should be possible to save the camera positions, and import the images with the correct (greyscale/iron) pallet for the creation of the orthomosaic, once the alignment is successful?

Is there any help, in using an existing DEM og GCPs in the area where the thermal data has been captured?

If possible, i would like a quick talk with you via skype or similar. I dont know if you offer this service. Finding a solution for this issue i rather important, as is has a large impact on the way we collect data/plan flights.

Kind Regards
Sune

[Alexey Pasumansky]

Hello Sune,

You can try processing the set with the smaller step between the images.

Side overlap is not required, but it may improve the stability of the camera alignment. But I don't that two synchronized cameras mounting side-by-side can be considered as two flight lines, due to the negligible stereo base.

The image resolution is quite important, usually we recommend to process at least 3-5 MPix images, but for thermal data the common resolution is usually even below 1 MPix.

If you modify the color/contrast for the images, but keep the same naming convention, you can substitute the image set using Change Path option from the Photos pane context menu for the orthomosaic generation.

The external DEM in the same coordinate system can be imported to be used as a surface for the orthomosaic generation (photos  would be projected on this surface). It may be helpful, since the DEM from the thermal data may be quite noisy.

[Sune]

Again, thank you for the swift respons Alex.

I sent a PS project via wetransfer. Its a single pass, captured with the 640 sensor (13mm lense).

No refference file has been used. No camera calibration added. Default values/settings in all windows.

Would an external DEM help with the creation of the orthomap. The point cloud looks very good, but the map is twisted.

Any comment on a skype call next week to clarify things? I have more related questions i need answered, before we may want to buy PS for use in our work.

Kind Regards
Sune

[Alexey Pasumansky]

Hello Sune,

Another issue with straight linear flight may arise due to the rotational degree of freedom around the flight direction. To avoid that you'll need either to use the GCPs or to load yaw, pitch, roll information for the camera orientation to the Reference pane.