HOW TO TAKE PICTURES FOR PHOTOMODEL3D

Inferring 3D structures from 2D photographs is considered an “ill-posed” problem in computer vision in the sense that its solution is inherently under-constrained and unstable. Visualsize's Photomodel3D employs an analysis principle called “structure from motion” for such a 3D inference process, and the process comprises many intertwined components. Failure of any of the components can defeat the whole modeling procedure. Hence, to ensure success, some caution should be exercised in gathering pictures for the 3D analysis.

What Types of 3D objects Can Be Modelled?

Please take a look at our demo pages. While our Photomodel3D program works with many different types of objects, the important thing to remember is that it does not work well with objects whose appearance is too plain (a single, or a few, colors with no texture), shinning (glass and mirror), or changing (moving or deformable objects).

Basic DO and DON’T

# Do take a lot of pictures (5 at least)

# We recommend a spatial resolution of 2M to 3M (e.g., 2048x1536, 1600x1200, and 1280x960 pixels). Do not use a spatial resolution that is lower than VGA (640x480).

# Do make sure that adjacent pictures show significant overlap

# Do make sure that the direction where the camera is pointing at does not change significantly from one photo to the next

# Do make sure that the object of interest occupies a large portion of the photograph

# Do not change any camera setting (zoom, focus, and aperture) when taking pictures, just point and shoot

# Do not stand at a single place to take pictures. Walk around the object of interest and take pictures from many different locations

# Do make sure that there is sufficient lighting so that the photographs show enough surface details

# However, do NOT use flash. If you need flash to take good pictures, then the environment is too dark for a good 3D reconstruction

# The most important rule of thumb is that the part of the 3D scene that you would like to model should appear in at least 3 adjacent photographs. The object of interest should occupy as large a portion of the photograph as possible. For example, to generate a 3D model of a figurine, your pictures should look like the one on the right (not the one on the left where the object is way too small) of the following illustration. You can use either portray or landscape modes.

While you can take pictures from random locations around an object of interest, it is harder to guarantee a uniform and complete spatial coverage this way. Hence, we recommend that you take pictures by moving the camera along a smooth trajectory (with small changes in the camera's position and orientation between adjacent shots). This is particularly important if you desire 360^o construction, the camera should move in a roughly “orbital” trajectory around the object of interest. You need to make sure that the viewing direction does not change too drastically from one photo to the next. Take more photos of narrow, curved parts of the object and less of large, planar regions. Generally speaking, you should take at least 30 pictures in a 360^o construction. Ideally, you should have around 60 pictures.

Special Instrucion for 3D Face Modeling

While human faces are not rigid, our 3D modeling software has had good success in creating 3D face models. However, it is important that one follows some simple procedures in taking pictures for building 3D face models:

# The subject should stay stationary (no laughing, talking, moving, giggling, etc.).

# There must be enough light (outdoor is better than indoor)

# Use the portray instead of landscape mode and make sure that the head occupies a significant image region. Click here for examples of good, big head shots in portray mode and click here for examples of bad, small head shots in landscape mode

# Use a simple orbital motion around the head from left to right (or right to left) and take about 20 pictures.

WHICH PROGRAM TO USE: 3D MODELLING OR MEASUREMENT?

3D model building and 3D measurement are two different products offered by Visualsize.

If you desire a complete 3D model of an object or a 3D scene - with dense surface structure and texture mapped surface description, you might be interested in our 3D model building software. This model building process is fully automated without the need of user intervention. But we do need a good number of pictures (from any consumer-market digital camera or phone, all without prior calibration and use of special photography equipment) to make it happen. We output the 3D models in standard formats that you can import into a CAD program. Then you can animate, visualize, measure, and modify our models as you wish.

If you desire just a few 3D length/angle measurements without building a complete 3D model (e.g., for home improvement projects, you may need to know just a few dimensions - the square footage of the roof or the kitchen counter, etc.) or you do not have access to a CAD program for post processing, visualizing, and measuring our models, you may try our metrology engine for 3D measurement. In this case, you need to provide only 2 photos from slightly different viewpoints, and then use our software to select features and dimensions for measurements. Certain manual intervention is required to tell us what objects in these photos that you want to measure.

So the difference is in the details of the models, the number of photographs required, and if you want to use our software for measurement and reporting, or export our models to another CAD program.

You can try our 3D measurement software directly from the Web for free. It is based on the software-as-a-service model. This link contains a tutorial on how to take pictures and how to use our software. It also contains a link to our program. We are in the process of setting up the 3D model building software using a similar SaaS model. Right now, please follow the instruction on this Web page to take pictures, and then email the photos to us (or let us know the URL to download the images). We will build the 3D models for you off line.