Model initialization for reconstructing models with datasets that contain variable accuracy GPS, like when combining aerial and terrestrial data where the terrestrial imagery may have lower confidence GPS than the aerial imagery.
The choice of initial camera pairs can drive the creation of the entire model resulting in wildly skewed models like this:
When the model when orthorectified should look like this:
Removing GPS data for shots with low quality GPS can result in much better models, but forces the use of BOW matching which can be much slower than OpenSfM's default incremental approach:
I propose adding an optional scoring factor to compute_image_pairs that prioritizes high quality estimates of GPS over low quality, when estimates are available
I have also posted about this on the community forum here:
https://community.opendronemap.org/t/reconstructing-models-from-multiple-levels-aerial-and-ground-initial-findings/11817
Model initialization for reconstructing models with datasets that contain variable accuracy GPS, like when combining aerial and terrestrial data where the terrestrial imagery may have lower confidence GPS than the aerial imagery.
The choice of initial camera pairs can drive the creation of the entire model resulting in wildly skewed models like this:
When the model when orthorectified should look like this:
Removing GPS data for shots with low quality GPS can result in much better models, but forces the use of BOW matching which can be much slower than OpenSfM's default incremental approach:
I propose adding an optional scoring factor to compute_image_pairs that prioritizes high quality estimates of GPS over low quality, when estimates are available
I have also posted about this on the community forum here:
https://community.opendronemap.org/t/reconstructing-models-from-multiple-levels-aerial-and-ground-initial-findings/11817