2011-10-10

2D to 3D with Photofly. Introduction.


 "Capturing the reality as-built for various purposes (renovation, rapid energy analysis, add-on design, historic preservation, game development, visual effects, fun, etc.) is now possible using your standard point and shoot digital camera thanks to advanced computer vision technologies made available through Project Photofly." - this is the official overview found at Photofly. But how good it really is?

The software is free, You can download Photo scene editor here (free registration needed). I've used point and shoot digital camera Nikon S8100 (4000x3000 pics). So here is my tests, results and conclusions:

TEST #1. First try.
169 photos, but only about 100 were automatically stitched (none of the close-ups were automatically stitched).


Sample picture:

Here you can see how many pictures were taken:

 And the result is: 
 

Conclusion:  
poor quality model, not very suitable for further exporting and modeling.

On Youtube comments I've got an advise to use less pics, that led me to Test #2.

TEST #2. Defining number of pictures.
Here is the same scene as in test #1, same photos, only less of them, about 30 (as advised in youtube comments).

Here you can see how many pictures were taken:

The result is actually worse then in test #1:



Conclusion: 
I think the main problem is the artificial lighting (6 lamps from top) and the glossy surface of the cubes. The amount of pictures taken doesn't play the main role, it's more important to make good quality pictures. So lets do the quality test.

TEST #3. Defining picture quality.
Clay figurine with raw, bumpy, matte surface. Same scene, different quality pictures.

Sample picture 1:



Result 1:



TEST #4. Defining picture quality.
Here is the same scene as in test #3, the difference is that I took slightly better photos.


Sample picture 2:



Result 2:

Conclusion: 
the better quality pictures You'll take, the better quality model You'll get. This conclusion is obvious, even without tests. What I've noticed, that PhotoFly does it's job with row, bumpy, mate surfaces, and just can't handle glossy surfaced objects. So If You need to "3D scan" a glossy and shiny object, choose lighting carefully, to avoid as many reflections as you can. I can make the presumption, that it's best to take photos of shiny and glossy surfaced objects in natural lighting, but without direct sunlight on them, or using professional studio lighting.

TEST #5. Complexity of the scene.
I've tested how Photofly "scans" objects, so I wanted to find out, how software works with more complex outdoor scenes.


Sample pictures:


Result:



Conclusion: 
The result actually impressed me. It seems not very realistic, but You can see how software captured and modeled even very complicated tree branches. Notice, that I made only standart pictures, not thinking about the best way for the software to recognize them.

In the Next post I will test Photofly capability to make a full 3D model of a real building. There will be some problems to solve, so make sure to follow new post.

No comments:

Post a Comment