This title appears in the Scientific Report :
2016
Please use the identifier:
http://hdl.handle.net/2128/13478 in citations.
Automated Soccer Scene Tracking Using Deep Neural Networks
Automated Soccer Scene Tracking Using Deep Neural Networks
Sport events exists in a vast amount of variations over the whole globe. While some events are privileged to gather a great audience by providing a television or internet broadcast, most of them could only be seen at the location of happening or amateur videos. The broadcasting of a sport event requ...
Saved in:
Personal Name(s): | Bodenstein, Christian (Corresponding author) |
---|---|
Götz, Markus / Riedel, Morris | |
Contributing Institute: |
Jülich Supercomputing Center; JSC |
Imprint: |
2016
|
Conference: | IAS-Symposium, Juelich (Germany), 2016-12-05 - 2016-12-06 |
Document Type: |
Poster |
Research Program: |
Data-Intensive Science and Federated Computing |
Link: |
OpenAccess |
Publikationsportal JuSER |
Sport events exists in a vast amount of variations over the whole globe. While some events are privileged to gather a great audience by providing a television or internet broadcast, most of them could only be seen at the location of happening or amateur videos. The broadcasting of a sport event requires two things: First, a infrastructure including cameras, servers and a channel, where the viewer can join the event. Second, a camera director or cinematographer that catches the important scene in the playground, helping the viewer at home to follow the event. While there exists companies that provide affordable broadcasting infrastructures, the camera director will always be a high cost / working time factor.Abstract We provide a fully automatic computer pipeline, that is able to select the scene of interest in soccer games. The approach is based on Deep Learning Techniques, or Convolutional Neural Networks (CNN). While CNNs shows state-of-the-art performance in image and voice recognition, object tracking and especially scene tracking, is less investigated. We show that those algorithms are able to handle the challenge, and exhibit how it could be integrated in a broadcasting pipeline, to increase the audience range. |