Hi,

I'm using the GPS data from the SRT files of videos in Metashape. I have RTK with centimeter precision.  However, when importing I noticed that the .srt file's longitude and latitude is rounded of after 6 decimals example: [latitude: 36.007811]. This basically creates only 1m precision.

When using RTK, the actual precision is much higher and I strongly request to add at least 2 decimals so that it can represent centimeter precision.

In the attached image you can see how the gps coordinates look "rasterized" due to the lack of decimals. The gps exif data for photos contains much higher precision, so my assumption is that the information is there, its just not written into the srt file and rounded/cropped off.

Can this be addressed in an update? It will lead to much higher precision for the photogrammetry community.

Thank you!
Stephan

example:

This is the first time I see someone doing drone photogrammetry with video. I use this feature with vehicle-mounted cameras, especially stereo cameras, but with a drone... I'm confused, and a lot of questions come to my mind.

Are you using that to produce GIS raster ?
What is the volume of data per hectar for a video flight vs a photo flight ?
Is it beneficial to have a oversampled overlap induced by video ?
Isn't it against productivity to lost camera orientation data for each picture by using video ?

LV_Forestry Posted at 8-15 14:11
This is the first time I see someone doing drone photogrammetry with video. I use this feature with vehicle-mounted cameras, especially stereo cameras, but with a drone... I'm confused, and a lot of questions come to my mind.

Are you using that to produce GIS raster ?

Hi,
I'm actually doing neural scene representation like Gaussian Splatting with it  (similar to photogrammetry but not polygonal), for that purpose its incredible useful to rather have more pictures/second to cover more angles than the actual perfect photo quality. The photos allow me only 0.7 frames per second on the M3E. So that's why I'm rather recording a video with a Mission planned in Metashape...

The higher precision the GPS, for the thousands of images extracted from the video, the better and faster is the alignment process in metashape and the angular overage of the subject instead of just single photos every 0.7seconds.

Troja777 Posted at 8-15 14:50
Hi,
I'm actually doing neural scene representation like Gaussian Splatting with it  (similar to photogrammetry but not polygonal), for that purpose its incredible useful to rather have more pictures/second to cover more angles than the actual perfect photo quality. The photos allow me only 0.7 frames per second on the M3E. So that's why I'm rather recording a video with a Mission planned in Metashape...

I'm interested in what you're doing because Splatting seems to be the future of 3D modeling. Not so much for accurate volumetrics or mapping but for 3D reconstruction in general. So are you using Metashape for the alignment step and then exporting those camera positions for your Splatting software? I have done exactly that in the past with NeRFs but I haven't found a way to do it yet with splats. Have you tried running the extracted frames through PPK processing? I'm not sure how/if that would work but maybe the .SRT file could be used in plack of the .MRK file like what would normally be done.

Johnnokomis Posted at 8-15 18:03
I'm interested in what you're doing because Splatting seems to be the future of 3D modeling. Not so much for accurate volumetrics or mapping but for 3D reconstruction in general. So are you using Metashape for the alignment step and then exporting those camera positions for your Splatting software? I have done exactly that in the past with NeRFs but I haven't found a way to do it yet with splats. Have you tried running the extracted frames through PPK processing? I'm not sure how/if that would work but maybe the .SRT file could be used in plack of the .MRK file like what would normally be done.

Hi,

there is a github page that has a gaussian splat exporter:
https://github.com/agisoft-llc/metashape-scripts/blob/master/src/export_for_gaussian_splatting.py

it works great! it exports a metashape alignment as a colmap format, which is what GS uses. I have not done ppk processing, not sure if you could post process a truncated GPS position?

Troja777 Posted at 8-15 18:43
Hi,

there is a github page that has a gaussian splat exporter:

The latitude and longitude in the .SRT file wouldn't matter and would be ignored. All the .SRT file here would be used for is the timestamp associated with each frame. The .obs file along with the base station file would then calculate the exact location at that given time for each frame. If each frame could be extracted with the exact filename that is listed in the .SRT file then it is definitely do-able.