Visual Recognition for Large-Scale and Weakly-Labelled Video Data

The main objective of this project is to investigate, develop and evaluate state-of-the-art computer vision and machine learning techniques, which are suitable for accurate modeling and recognition from large-scale video datasets that are weakly labeled. In particular, we will focus on the learning of visual recognition models for an application area of interest to SPORTLOGiQ Inc. – person re-identification for monitoring and tracking of player, activity recognition and group behavior understanding, and player and team performance evaluation in sports games. Learning recognition models in such cases typically leads to complex and ill-posed optimization problems, where video data sets are weakly-annotated. The recent years have witnessed substantial technical advances in areas such as deep learning (e.g., convolutional and recurrent neural networks), transfer and weakly-supervised learning, information fusion and distributed optimization, which promise to address such complex visual recognition problems, previously thought intractable. TO BE CONT’D

Faculty Supervisor:

Éric Granger

Student:

Partner:

Sportlogiq

Discipline:

Engineering

Sector:

Information and Communications Technology

University:

École de technologie supérieure

Program: