I’m using photogrammetry for erosion studies on trails. I carry the camera on a painter’s pole to obtain the data. In areas with complex shadows produced by tree leaves, the light and shadow patterns are converted to false topographic relief. Agisoft does not do this. Is there a fix in Pix4D?
To clarify, the gray images are hillshades, not the actual DSM or DEM.
Hopefully Pix4D or somebody else has a solution but as much as I love Pix4Dmapper, it isn’t the best software for every photogrammetry project.
I’m not sure what your work-flow is given these images.
Have you tried Cloudcompare to calculate cloud asymmetry?
This is a volumetric or (change detection) calculation that isn’t fully implemented in Pix4d yet.
I’m very interested to hear any suggestions from Pix4D on this as well. I have seen false topographic relief in DSMs/DTMs due entirely to pixel color differences as well (for example, white painted parking stall lines depicted as having up to 0.5 cm lower elevation than the surrounding unpainted asphalt, which is certainly not the case in reality).
If there are data collection or processing methods we can use to avoid or minimize these errors, or if Pix4Dmapper can be improved to eliminate this bug, it would be great.
Hi Gary, Thanks for the suggestion. I do all my elevation change work in ArcMap once I have accurate DEMs. And…Agisoft makes fine DEMs for me now. I’ve been hoping to move my workflow over to Pix4D desktop…but current version is not going to work, unless the tech gang can point me to a processing step I’m missing.
Data collection sets foundation for which all other processing and analyst rest.
All to often we see here poor sensor performance due to a number of reason.
To some degree in your case sensor performance I think caused these artifacts.
As pixel values approach either end of the sensors dynamic range a pronounced increase in triangulated errors tends to occur,. It seem especially evident at the white end.
Could you share the settings you’re using to process the datasets?
Here is the processing report. Does that have what you need Robert?
Doug Smith, thanks for sharing the report. I see the noise could be caused by a few potential reasons.
- The overlap is not ideal in the areas where I see shadows. Maybe adding manual tie points would help fix the geometry in the trouble areas.
- You have sharp selected for DSM type. Try smooth and compare results. I see similar noise in the agisoft hill shade; this may just be a matter of settings differences.
These would be my suggestions to try first. I would use the ray cloud and point cloud, add manual tie points and then re-optimize. The points will reflect the surface so viewing that as a form of QC will help you understand better how the surface will process.
Hope this helps, and look forward to seeing the results
Just to highlight what Robert said…that ATP picture needs to be white, not black. In Step 1 if I see even 20% black (yours is 90% black) then I stop and don’t even run Step 2 or 3. Either change the setup in Step 1 or re-take the pictures with A LOT more overlap if you want Pix4D to be more accurate.
Pix4D ran a solution for your data set but clearly the quality report shows that the results are totally unreliable.
Thanks Robert and Adam.
Will check on that shadow artifact for a dataset where I have more overlap…and will try the smooth v sharp filtering.
Could you also comment on the similar artifact posted by Pat Iampietro (above). Seems like a very contrasty change from black to white in a parking lot results in “topographic artifacts” that are not real.
Here is my processing report for the parking lot example I posted. This was flown with a SenseFly eBee RTK, so no external GCPs were used.
Lots of overlap, ATP map is nice and white. I also used sharp surface smoothing like Doug, maybe I should try smooth?.
Granted, the elevation differences I’m seeing from the white stripes are relatively small (a few mm), but they are pervasive, so if there is a way to avoid them I’d love to know how.
Thanks to anyone that can comment and to Doug for letting me hijack his thread!
I haven’t had a chance to go back and try “smooth” surface smoothing yet…
Am I expecting too much from Pix4D’s surface modeling?
Any thoughts appreciated.
Did you say a few MM on the white stripes? Could it be that the “White Stripes” are actually a few mm’s in elevation changes?
Also, may I ask what your end product is being used for? Do you know if you are in need of DSM or DTMS?