Tell Me the Truth

I finally got the non-commercial version of FaceAPI working inside OF.
Right now, the API runs inside an independent thread.  I’m working in windows on of my mac, and I hope that in a real windows machine the application started faster.

I’m using the OF version for visual studio,  to make it work with the 2008 version it’s necessary to remove GLaux.lib from the linked libraries. You can find more information and the link to a 2008 friendly version in the OpenFrameworks forum

Inside of the tools folder of the API there is a Camera Calibration Tool, CamCal.
The lens calibration files can be used for accurate tracking in world-coordinates.

Seeing machines specifications are  Intel Core-2 Duo, 2.4GHz, 4MB Cache, Logitech Quickcam Deluxe for Notebooks

For the IR setup I was planning to use a  security camera Sony SSC-M183 black&white CCD Camera with an infrared  filter  75mm Wratten-87C over the ccd and IR illuminators. Unfortunately  I’m not going to be able to bring the setup to Santa Fe because I’m missing the Video-to-FireWire converter.

Basic example of a 3d object rotation being controlled by the face tracking coordinates.

I’m excited about trying the Z camera in Santa fe and to be able to play with its software.
It seems like it could be an interesting option.
Meanwhile I’ve been reading about the possibility to use a stereoscopic camera.

I did a price request for a BumbleBee camera.  The BumbleBee seems to be a good option. It solves the challenging problem of alignment and calibration of both cameras.
It’s also possible to make it work with OF, but accordingly with Golan Levin experience,  although you can get disparity maps the amount of noise makes very different to implement a tracking system.
The price of the one that I was looking for (640 x 480 Bumblebee2) is  $1895 + $100 for the development kit.

OpenFrameworks is a free and open source application framework designed for creative coding.  OpenFrameworks is written in C++ and runs on Windows, Mac OS X, and Linux.  It is developed and maintained by Zachary Lieberman and Theo Watson and a growing group of collaborators.
The library is a multimedia C++ framework for artists, and on some level it attempts to do for C++ what Processing has done for Java.

In addition to the computer vision library OpenFrameworks also allow to load files exported by Blender. (One of Corey’s suggestion for the 3d part)
I’ve found some examples about the integration with blender and thought  they work with very simples geometries,  it is definitely a feasible option.

The tracking:

The red square determines the area in the video that the computer vision system is detecting as an human face.  This coordinates are the information that we can get from the program.
This tracking is bidimentional.

For the computer vision part of the project the idea is to use openFrameworks to implement the face tracking that is going to control de avatar.

Thanks to the openFrameworks forum I found more information about the algorithm used in the openCV library.

The method is based on the research of Paul Viola and Michael Jones published in the paper Robust Real-Time Face Detection, and extended by Rainer Lienhart.

Although openFrameworks was my first option to implement the project I’ve been also researching about different alternatives.

One of the most interesting is the new faceAPI of seeingmachines that apparently has a very robust 3D tracking solution and that has exactly the functionality that the project requires.

I’ve been in contact with Nick Cerneaz from Seeing Machines requesting information about a non-commercial license to use the API. Right now I’m waiting to hear from Nick Langdale to get more information about it.

Another possibility is the software that comes incorporated with the Logithec Quickcam Orbit MP
Not sure how “hackable” can be yet.