Can we transfer the 3D camera data into AutoCAD processable file?

[Yonghong]

My aim is to obtain a 3D CAD model for a room (wide * length * height = 5m * 5m * 3m) by using the RealSense R200. I just bought the R200, and have no idea how could I implement this task. As you may know that if we have a CAD file, we could then do some kind of design for the room, and decorate the room as beautiful as one like... Please help!

[Xusheng_L_Intel]

Scene Perception and 3D scan MW can generate obj file. It is standard obj file and can be used in AutoCAD.

[Yonghong]

Thanks David for the reply. The SDK is really great.

On the other hand, as it is said that the effective distance of the camera is no more than 3 meters, could you please give me a hint how can I scan such a big room? Do I have to scan one part of the room and save the obj file, and then scan some other part and so on. In the end, stitch all the different parts (obj files) together to form a whole room model?

[samontab]

You can merge together different scans, so you keep moving the camera, and the software stitches those 3D images together.

It is not a simple task though, have a look at techniques like ICP for example.

[Yonghong]

Thanks samontab. I have a few questions to ask you for help.

1. Have you tried to do the continuous scan before? How hard would it be to finish such a task (obtain a CAD model for a room of size 5m*5m*3m)?

2. What do you mean about "the software" that can stitch the whole room into one 3D image?

3. What does "ICP" means?

Thank you very much.

samontab wrote:

You can merge together different scans, so you keep moving the camera, and the software stitches those 3D images together.

It is not a simple task though, have a look at techniques like ICP for example.

[samontab]

1) Yes, I have merged together point clouds to represent a larger area than what a single one covers. It is a hard problem, you will not get perfect results every time. It depends, among other things, on the quality of the depth estimation that you can get, the algorithms used, etc.

2) The camera only provides depth snapshots of each area of the scene. Imagine this as creating a panorama from smaller images. The software is basically the hard part here, because you have to match different view points, and create a 3D model of the environment. This is by no means trivial, but is usually seen as trivial because our brains are extremely good at doing this.... actually having software to solve this is very hard, and is state of the art research.

3) ICP is iterative closest point, is just one of the simplest techniques for matching two depth images together. This is just one step of the entire pipeline, and just one example method.

3D is really fun!
 

[Yonghong]

Thanks, samontab. 

I understand that to stitch the video together is the most hard part in the whole process procedure. Is there any means that we can use to reduce the hardness, for example, to mark the wall with some color pen, i.e., to arrange some control points on the wall, or use some other tools to obtain better result?

P.S. I had a quick look at your website, and learnt that you are doing excellent in computer vision :-)

samontab wrote:

1) Yes, I have merged together point clouds to represent a larger area than what a single one covers. It is a hard problem, you will not get perfect results every time. It depends, among other things, on the quality of the depth estimation that you can get, the algorithms used, etc.

2) The camera only provides depth snapshots of each area of the scene. Imagine this as creating a panorama from smaller images. The software is basically the hard part here, because you have to match different view points, and create a 3D model of the environment. This is by no means trivial, but is usually seen as trivial because our brains are extremely good at doing this.... actually having software to solve this is very hard, and is state of the art research.

3) ICP is iterative closest point, is just one of the simplest techniques for matching two depth images together. This is just one step of the entire pipeline, and just one example method.

3D is really fun!