Titulo Estágio
Analysis of events in sports using a multimodal approach
Áreas de especialidade
Comunicações, Serviços e Infraestruturas
Sistemas Inteligentes
Local do Estágio
Maia
Enquadramento
Every show needs an audience and all audiences need entertainment. The player-spectator
dynamic is explored from public entertainment events in Rome and Athens, to modern
international events broadcast live to billions of spectators, and is now perceived as a
bilateral relationship, which is crucial for the success of their organization. .. Although
technological advances have enriched the experience of local and remote viewers, the
challenge of truly valuing and meeting the individual preferences and needs of diverse
audiences has not yet been fully met.
Watching the show while being physically on site has always been the most engaging way
to experience the event. Although the architectural configuration of the stadiums that host
live events, such as sports or concerts, has not changed much since the Colosseum, the
technology is now being used to make the experience of the fans much more rich and fun.
Nowadays, to attract and influence audiences of live sports broadcasts, a successful
annotation of the content is necessary so that it can be easily searched by users on
different viewing platforms or by the television channels themselves in their archives.
To achieve this, we need to use video and audio frames to identify patterns and events to
summarize the entire event a posteriori.
It is intended, therefore, with this project, in a first phase, to implement deep learning
techniques to find relevant data about a given context of a sport through video and
audio processing algorithms and in a second phase to use the result of this processing to
automatically annotate the main content in the form of metadata.
Objetivo
The objectives to be achieved with the project should be as follows:
- Investigate the state of the art regarding video processing and analysis to detect events
and patterns in a sport to be defined.
- Design the architecture of an ecosystem that allows the multimodal extraction of relevant
information about a multi-camera sports video and the corresponding insertion of it in the
video as metadata.
- Implement and integrate the various components of the ecosystem in order to obtain a
functional prototype of the ecosystem.
- Test and validate the ecosystem
The data sources are multiple video streams from different matches that are available on the company.
Data Description : Video streams of different football matches with annotated metadata describing relevant event highlights, and in different time periods (day and night)
Data Size: 400 Gb
Data Format : Unstructured, MPEG-TS
Plano de Trabalhos - Semestre 1
[S1, S6] - Preliminary studies. Analysis of the main scientific and technological solutions
for the processing and analysis of sports videos
[S3, S12]- Survey of requirements and definition of the architecture of the intended ecosystem
[S3, S12]- Writing of the dissertation
Plano de Trabalhos - Semestre 2
[S1-S6] Implementation of the various components of the ecosystem
[S6-S9] Integration of different components
[S9-S12] Test, evaluation and validation of the prototype
[S1-S12] Writing of the dissertation
Condições
O aluno terá acesso a posto de trabalho próprio.
O aluno será enquadrado no departamento de inovação e I&D da empresa e contará com a supervisão próxima de um orientador técnico e de um gestor de projecto. Não se prevê que haja uma integração directa do trabalho desenvolvido com outros componentes de software desenvolvidos pela empresa. Esta proposta está enquadrada nos desenvovlimentos do projecto europeu DataCloud (www.datacloudproject.eu)
Ainda será feita uma análise das tecnologias e linguagens de programação a utilizar mas ter-se-á em consideração as seguintes: Python, OpenCV, Tensorflow, C++
Observações
- Knowledge of object-oriented programming languages.
- Good planning, design and documentation skills (English)
- Advanced knowledge in Multimedia and Image Processing will be valued.
Orientador
Alexandre Ulisses Fonseca de Almeida e Silva
alexandre.ulisses@mog-technologies.com 📩