I am correlating two bands of the same sensor that are shifted by few pixels. I use for this the orfeo toolbox with the library otbcli_BlockMatching. Here after an example of command line I use:
otbcli_BlockMatching -io.inleft Band3.png -io.inright Band4.png -bm.minhd -10 -bm.maxhd 10 -mask.variancet 10 -io.out MyDisparity2.tif -bm.minvd -10 -bm.maxvd 10 -bm.subpixel dichotomy -bm.step 10 -ram 512
I get the disparity map representing the shift along X and Y directions, however I would like to understand the following issues:
1) Is the pixel resolution of the disparity map linked somehow to the bm.step? Having the bands at 10m resolution and using the bm.step 10, the maximum shift values I have is in between -1 and 1. Using bm.step 1, I have the maximum shift in between -10 and 10. SO I think that to have the disparity map at the same scale of the input image you have to rescale the disaprity map multiplying by bm.step. Is it correct?
2) Did you observe any difference in result producing the same disparity map in linux and Window? It seems to me that the disparity map produced in Window is mirrored along the X axis.
3) For the outlier detection I use mask.variancet 10, however the library produce a warning related to this mask but it continues to process till the end. Could you please explain?
4) My two bands are in .jp2 so I export in .tif using QGIS Gdal tool. However the library returns an error and does not continue. It seems reading in the blogs that there is still an issue with GDAL converted files. So I have to export again in .png, In this way OBT works. Do you know about?
thanks
أكثر...
otbcli_BlockMatching -io.inleft Band3.png -io.inright Band4.png -bm.minhd -10 -bm.maxhd 10 -mask.variancet 10 -io.out MyDisparity2.tif -bm.minvd -10 -bm.maxvd 10 -bm.subpixel dichotomy -bm.step 10 -ram 512
I get the disparity map representing the shift along X and Y directions, however I would like to understand the following issues:
1) Is the pixel resolution of the disparity map linked somehow to the bm.step? Having the bands at 10m resolution and using the bm.step 10, the maximum shift values I have is in between -1 and 1. Using bm.step 1, I have the maximum shift in between -10 and 10. SO I think that to have the disparity map at the same scale of the input image you have to rescale the disaprity map multiplying by bm.step. Is it correct?
2) Did you observe any difference in result producing the same disparity map in linux and Window? It seems to me that the disparity map produced in Window is mirrored along the X axis.
3) For the outlier detection I use mask.variancet 10, however the library produce a warning related to this mask but it continues to process till the end. Could you please explain?
4) My two bands are in .jp2 so I export in .tif using QGIS Gdal tool. However the library returns an error and does not continue. It seems reading in the blogs that there is still an issue with GDAL converted files. So I have to export again in .png, In this way OBT works. Do you know about?
thanks
أكثر...