• Background subtraction, human detection and image rectification code
  • sample image

    We currently release two main algorithms:

    • a background subtraction algorithm, that provides better performance than the OpenCV version;

    • a human detection algorithm dedicated to videos captured by static cameras. It successfully integrates a joint learning between foreground and appearance cues.

    See the human detection code webpage.

  • Face segmentation using Coherent Probabilistic Index Maps
  • sample image

    The software provides the source code and data needed to train and use a model for skin, hair, clothing and background color modelling and segmentation, according to Scheffler and Odobez, 2011. The code can also be used to reproduce the experimental results of this paper.

    See the Face Color model webpage for more details and download.

  • Motion2D: robust parametric motion model estimation software
  • sample image

    original sequence

    sample image

    sequence compensated using estimated motion

    Motion2D is a multi-platform object-oriented library for 2D parametric (constant -translation, affine, quadratic) motion model estimation. It follows the robust approach of Odobez and Bouthemy. It can be applied to the entire image or to any pre-defined window or region in the image. Thanks to its robustness, it can compute the dominant motion of a the image (or of a region) even if the image contains other objects and contents that do not fit the background dominant motion (in the example on the left, the sign and the car). Thanks to the use of a multi-resolution approach, large displacements can be estimated.

    Webpage of code and license (free for academic research).

  • Probabilistic Models: temporal topic models and more
  • sample image

    The web site proposes different methods for mining temporal data for recurrent patterns. These methods are un-supervised (there is no need to annotate data), generic (they can be applied to various kind of data), temporal (the timnig information is captured within the models), probabilistic (they allow for easy interpretation and extension), proven: they have been used for mining various data such as video and audio.

    See the following webpage for mode details