More 3D Scans
Next week will be our last in our training at KADDB. Our 3D Laser Scanner is now completed and we're adding the final touches. The scanner has undergone tremendous upgrades and is now performing way better than it used to last month.
To put it in short, during the past month we've managed to complete multiple viewpoint scanning and 3D mapping. The device can now automatically capture data from multiple views of the object and merge these sets of data to form a complete 3D model.
We've corrected serious bugs related to surface sampling. We've also programmed algorithms to compensate for inaccurate calibrations and instrument errors. Using these and some additional tweaks we've raised the maximum sampling rate to an amazing 6.66 samples/mm while cutting half the scan time!
The models we're producing now are extremely accurate and look amazing.
Take a look ...
The figures above show two point clouds of the Garfield toy. The cloud on the left is an example of what we were getting a month ago while the one on the right is an example of current scanner output. The new model is of much higher resolution and is completely noise free.
First of all, to help you visualize how our scanner now produces complete 3D models by scanning an object from multiple angles, have a look at these ...
The above models are for a violinist statue. They are actually point clouds, but are so dense that they look like 3D models even before reconstruction.
The cloud on the left has a lightning effect that will help you visualize what the model looks like. The cloud on the right is a special point cloud where data collected from each viewpoint has been represented by a certain color. There are four colors in this point cloud because it was formed by four different scans, each being from a different angle.
I've taken closeup captures on the violinist statue's face (which is smaller than 15cm x 10cm) to show you how accurate our scanner's resolution can be. The models below were constructed using only 25% of the maximum scanning resolution ...
One of the objects we scanned was an old wooden eagle statue. The produced model showed all object details including tiny carved regions ...
The high quality of the models we're producing reflect greatly in rendering.
Here are few shots:
In case you're wondering, that's a camel on the right ;)
The guys at KADDB were amazed by the results of this project. We've been told that there is a good chance that our 3D scanner will be moved to their central research department where it will be allocated larger funding and continue to be worked on.
I have decided to build a separate web site giving the in-depth technical details for researchers. I will also provide some of the models we've produced and explain few ideas that make our scanner better than many others. The website, however, won't be available anytime soon because it's by time I have some rest now ;)
To put it in short, during the past month we've managed to complete multiple viewpoint scanning and 3D mapping. The device can now automatically capture data from multiple views of the object and merge these sets of data to form a complete 3D model.
We've corrected serious bugs related to surface sampling. We've also programmed algorithms to compensate for inaccurate calibrations and instrument errors. Using these and some additional tweaks we've raised the maximum sampling rate to an amazing 6.66 samples/mm while cutting half the scan time!
The models we're producing now are extremely accurate and look amazing.
Take a look ...
The figures above show two point clouds of the Garfield toy. The cloud on the left is an example of what we were getting a month ago while the one on the right is an example of current scanner output. The new model is of much higher resolution and is completely noise free.
First of all, to help you visualize how our scanner now produces complete 3D models by scanning an object from multiple angles, have a look at these ...
The above models are for a violinist statue. They are actually point clouds, but are so dense that they look like 3D models even before reconstruction.
The cloud on the left has a lightning effect that will help you visualize what the model looks like. The cloud on the right is a special point cloud where data collected from each viewpoint has been represented by a certain color. There are four colors in this point cloud because it was formed by four different scans, each being from a different angle.
I've taken closeup captures on the violinist statue's face (which is smaller than 15cm x 10cm) to show you how accurate our scanner's resolution can be. The models below were constructed using only 25% of the maximum scanning resolution ...
One of the objects we scanned was an old wooden eagle statue. The produced model showed all object details including tiny carved regions ...
The high quality of the models we're producing reflect greatly in rendering.
Here are few shots:
In case you're wondering, that's a camel on the right ;)
The guys at KADDB were amazed by the results of this project. We've been told that there is a good chance that our 3D scanner will be moved to their central research department where it will be allocated larger funding and continue to be worked on.
I have decided to build a separate web site giving the in-depth technical details for researchers. I will also provide some of the models we've produced and explain few ideas that make our scanner better than many others. The website, however, won't be available anytime soon because it's by time I have some rest now ;)
11 Response(s) to "More 3D Scans"
Maher said:
Wow! Wow! i am impressed!
i am one of those peeps who are waiting for the website dude :D
Ghaith said:
lol, I'll make a note on this blog whenever the scanner's site is out :)
a different perspective said:
This is very interesting! Did you design an algorithm that combines the multiple shots from different angles? And, I'm guessing you need to tell it what the relative angles between the shots are.
It's amazing how the current output is noise free. It looks like a really fun project to work on.
Ghaith said:
Yes, the algorithm you're referring to is a simple mathematical translation that rotates the captured data points about a central axis of rotation.
We also know the rotation angle up to a great accuracy because our 3D Scanner includes a rotational actuator that actually does the rotation.
These two pieces of information are sufficient to merge multiple shots together, although practically you need some other measurements to compensate for instrument errors. In our device, all of this is done automatically.
Anyway, you're definitely correct regarding this being fun to work on :)
Zait o Za3tar said:
Very interesting. i did a course on computational perception (vision) and it was so interesting. Are you building the scanner?
are you using focal lengths, opening width and other parameters to determine the depth of the points??
Looks like a very promising work :)
Zait o Za3tar said:
I just read your previous post about the project! so I guess it answers my question :).
sharkooseh said:
lol jameel wallah
tayeb laish ma te3mal scan la 7alak? :P
Ghaith said:
Zait o Za3tar
Yes, I am building the scanner with two friends. The scanner uses laser to measure depth as you should have probably guessed :)
sharkooseh
I got asked the exact same question a couple of hundred times =)
The answer is simple; to scan myself I need to stay motionless within 0.1 mm accuracy for about 24 hours and I also have to give away my eyes because direct exposure to the laser beam will cause blindness.
I'll be happy to scan you anytime should you consider volunteering ;)
sharkooseh said:
laa! ba3dain ma 2a2dar ashoofak!
ma ba2dar 2a3eesh lool
bas ino DDDDDDDDUUUUUUUUUUUUHHHHHHHHHHH!!!!!!!! :P
Ghaith said:
Very funny stuff :)
ma3en said:
hey gaith u know? u remind me of myself 5 years ago! i dd a project with the same concept, so if u need any info send me :)
here are tips for u:
1) use FORTRAN or COBOL coz they're strong languages, if used another language, switch to any of them
2) when i used to debug a problem that i face, i start observing the behaivour not the code, notice the error first, then fix the code
nice work though and keep me updated :)
Write a Comment