Support Website Contact Support Blog

Processing historic thermal images without coordinates - all cameras projected at 90degr.


I have received 10 000 (640x480) thermal images captured with a FLIR Duo Pro R from a helicopter without a IMU unit that I need to orthomosaic. I have no image coordinates but I have a flight plan so I can manually identify locations in images and retrieve coordinates. However, doing this for 10k images is not realistic.

The following sensor model parameters worked with another dataset from another acquisition (with coordinates).

The flight plan is a south to north grid pattern.

For my first test I took 255 images flown in parallel so there should be sufficient front and side overlap.

Unfortunately, during pixel alignment step all the the cameras are all projected sideways. This is because the software ‘sees’ similarly looking pixels in images at the start and end of the flightplan. This results in a moon like shape and all images are tilted 90 degrees.

My first idea was to find the coordinates for 5 images at the most northern and most southern part of the flightplan. I assume this might prevents the cameras from matching pixels in images that are at the opposite end of the flight line.

Anyone has any tips or ideas how to tackle this?

test_espoo_2May_parrallel_lines_unreferenced_report.pdf (956.7 KB)

Flight planEspoo_flightplan_may2

Hi @subscriptions, If you have timestamps of when the images were taken and in the trajectory, I would try and match these together to provide rough coordinates. You might also need to force Pix4D to use these in the calibration method, however, it is definitely not guaranteed to work.

If you need any help doing this please let me know at

Hi Jason,

Unfortunately, I don’t have any trajectory data. All I have is a report with a flight plan image. Your suggestion did work earlier for me as well with another dataset for which we did have IMU trajectory data.

I I’ll just try and see if adding a few coordinates to some images might do the trick.

Would it be possible to obtain this information from the helicopter flight records? I think that would probably be the easiest way. Failing that, I think you would be better off creating GCPs/Tie points on the ground, rather than adding image position data.