First look the tool is impressive. My test case is a handful of google street view (GSV) image screencaps (just messing around). Very sparse as I’m sure you’re familiar with the way GSV works.
I got an initially promising result with a mesh roughly the size and shape of the focus building, but was unable to get a different view. There is one green frame and one blue frame in my image set. I have 13 images, can see 2 sides of a corner building.
I expected all images to be processed?? I can see 3 cameras in the 3d model view space.
I went through the quick start but didn’t see a minimum images/recommended overlap, other than my image set is way, way sparse.
Anyone have experience with street view to get urban reconstructions?? At this point I’m just interested to know I’m using the tool right, but also would like to know more about how to make the results better.
Image attachments not supported??
Thx
Hi Shawn Allen
the GSV workflow can work, but need a plenty of imgs ( overlap ) to get “decent” model quality…
can post screenshot so can better understand what you deal with ?
I’ve put an image online as well as my project, as I suspect that might help.
Thanks for checking it out!
https://www.nads-sc.uiowa.edu/userweb/sallen/capture.html
Image:
Hi Shawn Allen
yes as can see RC (and other sw too ) are not good for so low images count to get useful results
when have very low images then its better to use IMAGE BASED modeling
if want use photogrammetry, then need just much more image so it can reliable work
Ah.
Would a video source be sufficient, or do the images need some angular differentiation in order to work?? I’m guessing the answer is no, and yes.
For example, driving a car down a road, taking a video, extracting images every 1-2 seconds would produce a pile of images.
Meanwhile I’ll go poke at the FAQ.
Thanks!
Hi Shawn
Would a video source be sufficient, or do the images need some angular differentiation in order to work?? I’m guessing the answer is no, and yes.
this is named as forward looking capture its not the best one BUT if could get say 2 cameras and sync them (in msec range ) for picture capture and set them on say both sides of car ( stereopair config ) and make them look on same “target” say 5m in front it would be the very good setup.
the Videogrammetry ( using video for recon ) is not very good if cannot have RAW video data as the video CODECs downgrade the image quality ( sometimes very strong ) + you still have just 4k video (3840 x 2120 pix ) approx. then if use it for recon you get NOT per pixel 3D information, you get say 1/4 to 1/8 of resolution 3D results. So do the math what is result resolution…
best if use “classical” photogrammetry workflows with say 12+ Mpix camera
