CRISP: Sport gesture recognition using Deep CNNs applied to Table Tennis

Coordinators:

Renaud Péteri - Université de la Rochelle
Jenny Benois Pinaud - LaBRI
Julien Morlier - Director of the sport faculty of the Université of Bordeaux


Other team members:

Pierre-Etienne Martin - PhD student - LaBRI - Université de Bordeaux
Alain Coupet - Senior Lecturer at Sport faculty - Université de Bordeaux
Ronan Le Merrer- System Engineer at Sport Lab - Université de Bordeaux
Mathieu Dubos - Bachelor student - Université de Bordeaux
Léa Godallier - Bachelor student - Université de Bordeaux


The project is supported by Region of Nouvelle Aquitaine and Bordeaux Idex Initiative


Start date : May 10th 2017
End date : Jully 10th 2023


Description :

The goal of the project is to increase sportsmen performances by using Digital technologies. Sportsmen actions are recorded in an ecological situation of their exercises and games.
The first targeted sport is Table Tennis. The goal of automtic analysis is to index video recordings by recognising table tennis strokes in them. An ecological corpus is being recorded by students of Sport Faculty of University of Bordeaux, teachers and table tennis players. The corpus is annotated by experts via crowd-sourced interface. The methodology of action recognition is based on specifically designed Deep Learning architecture and motion analysis. A Fine-grain characterization of actions is then foreseen to optimize performances of sportsmen.


Dataset :

Samples of our dataset per classes


Recordings using GoPro cameras
Speed recordings : 120 frames per second
Annotation interface




Data processing