Cheap 3D scanning for engine swaps

So I've been playing around with Xbox Kinects and all sorts of 3D scanning software. It's surprisingly accurate. It's not a FARO arm or NextEngine laser scanner, but I'm consistently measuring errors smaller than 1/8" over 36".

The next step is of course powertrains and engine bays. I think this is an incredible tool for visualization and project planning. If no one has done it yet, I'd love to start a website where people can freely share their 3D scans of cars and parts.

Click for video: The ugly beat up dull Exocet bodywork in the background of the video is our mockup set, our production bodywork doesn't look like that!

All you need is a Kinect and a computer. The easier software I've been using is ReconstructMe Console. There's a 3mx3mx3m setup that you can tweak. http://reconstructme.net/

Meshlab will let you open and clean up the files. I'm still learning how to use it and transfer cleaned up point clouds into SolidWorks. http://meshlab.sourceforge.net/

That is amazing! What kind of file format does it save it as?

J

ReconstructMe poops the data out in .ply format. Meshlab will take that and export a file compatible with just about any 3D modeling software. I use .stl the most.

DUDE, you are a genius.

buy a domain name quick!

catch 123d from autodesk does the same thing, no kinect camera, just a normal camera.

In reply to slopecarver:

Catch123d does, do something similar, but the Kinect's well-documented stereovision leads to much better dimensional accuracy. The AutoDesk method doesn't account for absolute scaling, either. Catch123d is good for making artwork, whereas the Kinect methods can be used to measure and design.

According to a few academic papers, the 123D algorithm is limited to an error of 1:500 for normal-vector deviation. If you take one shot every ten degrees, that error increases to 1:300. The Kinect supposedly does 1:1200 in a single capture and some algorithms have pushed accuracy up to 1:10,000.

Part of the advantage of using the Kinect is the consistent hardware. The lens distortion, spacing, focal lengths, etc of the Kinect are very consistent from Kinect to Kinect, and so the community has been able to develop methods to rival the accuracy of the entry-level ($1k) laser scanning technologies.

Cool, I've used the DAVID laser scanner technique, but I've always wondered about that Kinect sitting in my living room...

That's fantastic!

This thread has some examples of scans of an early RX7 and their use in mockups. It's pretty neat stuff.

Example:

In reply to MrJoshua:

That's sick. I knew I was far from the first to have the idea.

good E36 M3

Just bought a Kinect off the flea after reading this thread though I run Linux rather than windows. I'm putting a T5 engine in a '72 BMW 2002 touring so it should come in plenty handy!

Did anyone else make a similar move after reading this? Did the OP put up a website for people to share scans in the end?

Love the forum,

Matt

Is the Kinects moved in relation to the target or is the Kinects stationary and the target rotated? Can multiple scans (car body scanning) be stitched together?

Myself and a friend did this with a Dodge Rampage... we may or may not have a solidworks model of it with a Pikes Peak legal cage and a mid mounted, twin turbo V6 now.

The hardest part is working with the mesh once it's been exported- the next step is getting some sort of feature recognition going to simplify large portions of the scanned model.

RIP jhaas

Apis_Mellifera wrote: Is the Kinects moved in relation to the target or is the Kinects stationary and the target rotated? Can multiple scans (car body scanning) be stitched together?

The Kinect moves, and scans can be stitched but it's easy to end up with a HUGE file very quickly.

We have a Faro room scanner here at work. Check out youtube for "Faro Focus 3d" None of the videos are ours or anything. But the nice thing about the Focus 3D is the easy ability to scan multiple times and register them together.

Warren v, we have a couple of guys that will come to you and scan stuff(you can PM me, sorry for the shameless plug).

I think I still owe Keith a scan of a Miata.

I am very very interested. Not to take that pesky engine out to measure... m

Man I've been keeping an eye on this stuff but I didn't realize it had become this easy! Gonna watch out for a used Kinect now...

You had me at BMW 2002 touring.

Warren v wrote: The next step is of course powertrains and engine bays. I think this is an incredible tool for visualization and project planning. If no one has done it yet, I'd love to start a website where people can freely share their 3D scans of cars and parts.

Grabcad is the already-existing solution, but its a storage place (with some social aspect) and not much more.

I actually started a project like this a while ago, but shut it down. If you are looking for any help, I'd love to be involved. An open source enthusiast car part community other than LocalMotors would be awesome.

When I clicked this thread, I first expected to see this: https://www.kickstarter.com/projects/occipital/structure-sensor-capture-the-world-in-3d

I'm going to have to pick up a Kinect and give this a shot.

There are plenty of linux-based programmes to manipulate the meshes once scanned but i'm struggling to find one that just lets me 'use' the Kinect - do you have any suggestions? There are also plenty of tutorials about how to install it linux but not alot for the bit in between installation and playing with 3D scans!

Warren v wrote: ReconstructMe poops the data out in .ply format. Meshlab will take that and export a file compatible with just about any 3D modeling software. I use .stl the most.

How do you use the .stl? All the cad software I have used will not let you actually USE a .stl file - you can see it, scale it, rotate it, but not pick points from it, create planes, or otherwise "engineer" from it. It looks like all of the Meshlab export formats are some form of Mesh (fitting, I know), which always has this limitation in the engineering CAD world.

Edit: 3d dxf/dwg? Is that what you use?

Can someone clarify? What are the advantages of doing this? Is it so you can see how well an engine would fit into a particular engine bay or something?

In reply to Desmond:

Not just fit, but to help visualize any aspect, not just engine, but anything.