I'm trying to learn photogrammetry for VR and AR, mainly want to build environments to walk around in, and models to drop on a table in AR, and would love your advice on getting Reality Capture up and running --
I've read through a lot of the posts here trying to figure stuff out, and went through what tutorials and videos I could find ---
I probably started out too ambitiously with a building that was too big to fly loops around with the drone, so we scanned one side at a time, up and down, up and down then curving around the corners and over the top at 45' degree looking down.
I've now got it into Reality Capture, trying several passes with different settings, detector sensitivty on High or Ultra, bumped pre-selector features up to 25,000 or 40,000, back and forth between merging components=true and false, and at some pont went back and re-processed stills from the drone interval at 1 second from 5 seconds when I thought adding more pictures would help.
Now i'm at the point where I've gotten 1 component with 1033 pictures in it, but there are big chunks of that component that are askew, as hopefully you can see from the photos below:
At the bottom I've put an image from google earth that shows generally what it's suppposed to look like, so you can get a better idea for how the point cloud is off -- I've tried lassoing the areas that are off the charts, i.e., one chunk of the roof and a side of the building is shooting off into space above the building, and part of the side is a good 25 feet off the place where it should be --
1. Should my alignment settings change in some way?
2. What's the general theory for control points, i.e. for a building like this, how many control points do you think I need to use, and how many differente images do I need to get that control point attached to?
3. Since this one component is all askew, can I fix the realignment within this component with control points, or are control points really only helpful in merging different components together?
Thanks in advance for any pointers, really appreciate it!
Please sign in to leave a comment.