Section: Contracts and Grants with Industry

European and National Projects

FUI 3Dlive

Participants : Frédéric Devernay, Matthieu Volat, Sylvain Duchêne, Vijay Ch. A. V..

3Dlive (http://3dlive-project.com ) is a collaborative project, supported by French Ministry of Industry, and involving 3 industry and research clusters: Images & Réseaux (Brittany and Pays-de-la-Loire regions), Imaginove (Rhône-Alpes region), Cap Digital (Paris region).

There are eight partners:

• R&D/industry:

  • Orange Labs (project leader),

  • Technicolor (3D R&D),

  • Thomson Video Networks (encoders),

  • Thales Angenieux (optics).

• Small companies:

  • AMP (TV shooting),

  • Binocle (specific 3D HW & SW manufacturer).

• University labs:

  • INRIA/PRIMA,

  • Institut Telecom.

The objectives of this project are to create expertise in France for the live filming and transmission of 3D stereo contents, and to help French industry and universities to be major global 3D actors.

The role of PRIMA within this project is to develop new algorithms for real-time processing of stereoscopic video streams. This includes:

• stereoscopic video rectification and geometric adjustments.

• view interpolation, and extraction of stereoscopic metadata for the adaptation of the stereoscopic content to the projection screen.

These algorithms rely on view- and scale- invariant feature extraction, feature matching, dense stereoscopic reconstruction, and computer graphics techniques (matting, and accelerated processing and rendering using the GPU).

3Dlive won the Loading the Future trophy from the Images & Réseaux cluster in 2011.

OSEO Project MinImage: Embedded Integrated Vision Systems

Start Date: 1 march 2008

Duration: 60 months

The consortium consists of:

• STMicroelectronics

• Saint-Gobain Recherche

• CEA-LETI and LIST

• Varioptic

• INRIA Grenoble Rhone-Alpes Research Centre

• DxO

The goal of the MinImage project is to develop integrated micro-cameras for portable telephones. This is a 141 Million Euro development program provided with 70 Million Euros of Aide by OSEO/AII. The program includes major development efforts in micro-electronics, optics, image processing, and image analysis.

Within the MinImage program, PRIMA has created a fast integer-coefficient O(N) algorithm for computing scale and orientation normalized Gaussian derivatives that is suitable for implementation as a dedicated image processing component within an CMOS integrated vision system. The PRIMA feature extraction engine is currently under evaluation for use in the next generation integrated vision systems for mobile devices sold by ST Microelectronics.

Within MinImage, we have achieved video rate calculation an image pyramid with exactly scale invariant impulse responses using an integer coefficient O(N) algorithm suitable for embedded computer vision. Our software implementation software provides a practical method for obtaining invariant image features from very large retinas for detection, tracking and recognition at video rates. This method is at the core of the real time embedded image description system for mobile applications being developed by ST Microelectronics and the CEA.

John-Alexandre Ruiz-Hernandez has recently demonstrated that the steerable scale invariant Gaussian derivative features outperform the popular "Integral Images" method for face detection using a cascade of linear classifiers popularized by Viola and Jones. We are currently extending these results other applications such as gender recognition, character recognition and place recognition. Key results in this area include

1. Fast, video rate, calculation of scale and orientation for image description with normalized chromatic receptive fields.

2. Real time indexing and recognition using a novel indexing tree to represent multi-dimensional receptive field histograms.

3. Robust visual features for face tracking, bodies, and other objects.