Results

The i3DPost Project officially ended on 31st December 2010, however the work continues. Some results from the project can be seen in the 'Experimental Production - The Midas Touch'. Each of the partners had a hand in the outcome of this short, from providing plugins to actually doing the post-production. Thanks has to be given to the user partners (BUF & QD) for their time spent on it.

There is a showcase DVD that has been produced whose overview of the Project outcomes can be viewed here:

Some of the stages are represented here...

On-set Performance Capture

Set capture - Digital stills

Matting the Principal Camera

Semantic Labelling - detecting and tracking people

Temporal Consistent Mesh

Facial Performance Capture

The Foundry

As part of the work for WP5 Structured Face Modeling, the year 2 results from The Foundry can be seen below.

Dense dispartiy estimation for narrow-baseline video sequences has been extended to relatively wide-baseline multi-view camera configurations. Wide-baseline matching of SIFT features from research work on 3D Feature Analysis has generated camera-to-camera correspondences to build up the facial surface. The results are significantly more robust, which creates better facial surfaces.

Multi-view camera calibration and point cloud estimation for the face has been combined with KLT feature tracking to create 3D surface trajectories. This has been used to prototype a virtual marker tracking system for facial animation. A set of "virtual markers" are defined in the video sequence and tracked to create a moving 3D point for each marker.

University of Surrey

The test data sets were produced (D3.1 & D3.3). If you would like access to the data sets, they are available here See examples below.

an overview of the UniS work can be seen in this short movie...