3D scanning and modelling

Some of our 3D research is already documented in this post here, In this post I will try to cover all the other things not mentioned already.

We have been investigating ways to generate 3D content for use in virtual reality and augmented reality experiences. This has covered both 3D modelling, and 3D scanning.

Maya

I went through this tutorial, and tried to apply the same workflow to create a more organic looking model. This is the nicest thing I came up with :

Here's the same model imported into TouchDesigner :

Z - Brush

In terms of 3D modelling, there is an awful lot of software out there, but one of the nicest things I came across was something called Z-Brush. Z brush is unlike most other 3D software packages out there, in that it takes a very physical approach to modelling. You get what is essentially a little ball of clay to start from and push, pull, crease, and smooth it to get the shape you want. This is a lot more natural than manipulating points, vertices and faces, and allows for much more organic shapes.

It is mainly used for creating extremely detailed, organic models such as those for video games, concept art or CG movies.

My first attempt at making something in Z-Brush turned into something resembling a bat

                            

I introduced it to some of the young people at the Self Managed Learning college, where David, Louis and Chris teach each week. They took to it extremely quickly, and made everything from the silly and immature :

...to some really rather nice abstract compositions

(all of which I like much more than anything I have made!)

I quickly realised that recreating toy parts from scratch in Z-Brush was not quite as easy as all the fancy time lapse videos on youtube would have you believe. It really is a skill that you would need to develop over several years to get anywhere near this good :

So instead of creating models from scratch, we decided to look into how we could digitize existing toy hack creations, and find ways of 3D scanning them.

123D Catch

The most promising looking software I could find to do this is AutoDesk's 123D catch. It creates 3D models from a series of overlapping photos by matching points between images and measuring how far they move in relation to each other between shots. It all happens automatically, and you get a textured 3D model.

This all sounds great, and mostly it is, but as I mentioned in my previous 3D research post the models it gives you are :

 - very complicated  - (millions of polygons)
 - noisy, bumpy and innacurate unless you have perfect lighting conditions
   (this is even more pronounced on smaller objects such as toys)
 - often full of holes - (not good for 3D printing)
 - nearly always come with multiple texture maps  -  
  (makes it tricky importing them into game engines and apps like Aurasma)

I put all of the toys that we 3D scanned up on the 123D catch site, and you can see them in this gallery here

I think my favourite has to be "Pyramid Man" (I wasn't feeling particularly creative when naming my uploads)




I spent a long time going through tutorials to try and help me figure out how to clean up, simplify, and export the models it gave me. Being completely new to all 3D software I found it very tricky to navigate all the new 3D jargon people were casually throwing around. Quite often in a tutorial, you'd be following along, and they'll say something like "and now just take that into Maya and remap all the UVs", cut back to after it's done and you'd spend another two hours going through UV mapping tutorials trying to figure out what they did.

Here are a few of the best tutorials I found for doing this (although inevitably one step somewhere wouldn't work in each one and you have to piece bit together from other places to carry on)

Eventually I did have some success with Z-Brush, and discovered several handy tools such as the decimate tool (for intelligently reducing the number of polygons in your model) There is also a tool in Z-brush called Dynamesh, which can help turn this:

Into this : 

This model has been simplified using the Decimate tool, and is made of triangles rather than quads: 

Unfortunately once I had managed to simplify the model, getting texture information to then map correctly to the low poly model is something I never quite figured out. This was about the closest I managed to get :

This tutorial for Maya looks extremely promising, butt again I had issues with texture not appearing correctly

123Dcatch_Maya_Integration_Tutorial_part1 from Mike Haas on Vimeo.