trying to put together a video--> mesh workflow
And I am neither autistic enough to persist, nor ingenious enough to find the issue
My classmate from the uni coded a 3D app
Video-to-3D
yeah,i have the same issue; I had some luck earlier breaking the 3D PC down into layes of 2D PC
It kinda worked
But not really
Basically it was an app that approximated a spinning object in 3D
now need to convert in order to apply Open cv
Sometimes it worked, and sometimes just went horribly wrong
the idea is to take a video and use photogrammetry to create a 3d object
And with each revolution it would fuck up somewhere else, while fixing one spot
NNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNNIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEEELLLLLLLLLLLLLLLLLLLLLLLLLLLLLLLL
FFFFFFFFFFFFFFFFFFFFFFFFFFFAAAAAAAAAAAAAAAAAAAARRRRRRRRRRRRRRRRRRRRRRRRRRAAAAAAAAAAAAAAAAGGGGGGGGGGGGGGGGEEEEEEEEEEEEEEEEEEEEE
cause 2D sections<plane> are no different that 4D frames <images>
3D imaging from 2D has some serious issues with shadows as it turns out
it does, but if you strip the back ground
A shadow makes it thing that the spot is deeper/shallower than it really is
it works a ton better if youu use OpenCV to detect countours and then strip everything but your target object
It works better when the camera is monochrome and the object is lit by an array of lasers
then run those frame through the SIFT algorithm
lidar is best
but you can strip the color out of the image with one command with opencv
A parrallel course to mine was making one such table thing that digitised phyiscal objects
lidar actually has a similar problem; coordinate systems
They got some serious backing from bosch
Like they not just used a laser array and a dome of monochrome cameras, oh no that was too simple
Those lasers also measured distance each
yeah, I'm trying to integrate those two workflows so i can use photogrammetry info in cad and then laser data combined to photogrammetry
but everything uses different coordinate systems which aren't readily connected
In the end they abandoned the dome idea and went with a turntable that had an arch of cameras and lasers
Actually not an arch, more like almost a complete circle
a lidar array is the best way to go if you can afford the equipment
high accracy
it's just stitching such large datasets can be a bear
that's another area not connected; photo panaroma's and point cloud registration
One side contained lights and normal cameras (for colour data) and the other side had 3 horisontal laser lenses with an array of monochrome cameras and distance measuring inhole camera-chips
meme
MEME
there are some kick ass open source tools but they are all insulated from each other
