Partilha de Ciclos de CPU para Identificação e Indexação Facial em Multimédia
Duration: from 2010 Mar 01 to 2013 Feb 28
Financed by: FCT
Prime Contractor: INESC-ID (Other)
We will design a peer-to-peer cycle sharing infra-structure to support video analysis and indexing. Video processing, extraction, face detection and identification will all scheduled as embarrassingly parallel Bag-of-Tasks (i.e., no coordination needed among concurrent activities) by reducing the footage resolution, splitting video into smaller sequences, and scattering them among the nodes of the peer-to-peer overlay for processing. Intermediate and final results are stored and indexed leveraging layers and dimensions of the overlay.
PROSOPON will also support users' contributions to guide the process of face classification. The peer-to-peer infra-structure will be used with an additional overlay for distributed fault-tolerant storage of the video library and the index that results from the face indexing process. This additional overlay index will have a multidimensional structure according to the parameters used in the face identification phase (e.g. eigenfaces/eigenfeatures) and will be used to efficiently execute queries on the indexed library.
Project Web Site: http://gsd.inesc-id.pt/~lveiga/projects/prosopon
Partnerships: INESC-ID (Other)
Principal Investigator: Luis Manuel Antunes Veiga