Hi all,
Context
*I try to register two Pleiades images that come from an orthorectification process and that are around 3m offseted. I’ve tried 2 methods : Residual Registration workflow, and FineRegistration. One raise an error and the other works but gives an ugly result. *
Configuration setup
My system: Ubuntu 22.04
Version of the OTB: 9.1.0
I installed the OTB with: the binaries
Description of my issue
1st case, using Residual Registration workflow :
The error raises at otbcli_RefineSensorModel step, here is the command and the error message :
otbcli_RefineSensorModel -elev.dem DEM/ -elev.geoid Geoid_EGM08/EGM08_2154.tif -in "slave_im1_superimposed.tif?&geom=TheGeom.geom" -out refined_slave_image.tif -inpoints homologous_points.txt -outstat stats.txt -outvector refined_slave_image.shp 2024-11-26 14:13:59 (INFO) RefineSensorModel: Default RAM limit for OTB is 256 MB 2024-11-26 14:13:59 (INFO) RefineSensorModel: GDAL maximum cache size is 3205 MB 2024-11-26 14:13:59 (INFO) RefineSensorModel: OTB will use at most 64 threads 2024-11-26 14:13:59 (INFO): Loading metadata from external geom file TheGeom.geom 2024-11-26 14:13:59 (WARNING): The SensorTransform factory could not find a compatible Sensor Transform 2024-11-26 14:13:59 (WARNING): The SensorTransform factory could not find a compatible Sensor Transform Segmentation fault (core dumped)
Thinking it could be a memory issue, I’ve tried to reduce the size of the input image but I get the same result. Thus I’ve no more idea of what could be wrong.
2nd case, using FineRegistration :
I do first a Superimpose of the slave image on top of the reference one. Then I apply the FineRegistration. I’ve played a lot with the parameters erx, ery, cox, coy (side question : if we give cox and coy, can we lower the erx and ery values? what is good practice with these parameters?) and the result is well registered but … the radiometry is ugly. It looks like a wrong resampling. Moreover, there is a residual fringe around some features. I put below the slave image and the warped one.
Best,
Jules