Convolutional Neural Networks for Teamsport Player Re-Identification.
Context: Democratic and personalized production of multimedia content is one of the most exciting challenges that content providers are facing today. In the context of teamsport production, Keemotion (www.keemotion.com) is already changing the way content is provided. Through the use of AI and patented algorithms, Keemotion's automated production system films and live streams high-level sporting events from all over the world without any human interaction. Typically, a set of cameras is used to build a panoramic view of the scene. Image analysis and game interpretation is then exploited to decide how to crop the panorama to show valuable content to the end user. The system provides a practical solution to cover local events at low cost, as no technical team or cameraman is involved anymore. It also promotes personalization of content, since knowledge about the content of the scene can be exploited to adapt and personalize content summarization to the individual user needs.
PhD: Advanced interpretation of a teamsport competition requires accurate identification of players all along the game. Off-the-shelf multi-object tracking solutions succeed in building reliable but partial tracklets through short-term association of detections in consecutive frames, but generally fail in tracking each individual on the long term, due to complex interactions between players. The proposed PhD aims at developing solutions to temporally associate the tracklets that correspond to the same player, to achieve long-term player identification. In practice, for a given query tracklet, the research will aim at estimating, among previous and subsequent tracklets, the ones that likely correspond to the same player than the query tracklet. Hence, it proposes to formulate the association problem as a player re-identification (Re-Id) problem. The specificities encountered in the teamsport context (similar appearance, but limited number of players on the field) motivate the design of original CNN-based Re-Id methods.
Co-tutoring: The PhD program is funded by Keemotion, and involves co-tutoring from EPFL (https://www.epfl.ch/labs/vita/) and UCLouvain (https://sites.uclouvain.be/ispgroup/). This means that the PhD student will be supervised jointly by Prof. Alahi and Prof. De Vleeschouwer, to be awarded a double-badged PhD degree by EPFL and UCLouvain. The candidate is expected to spend one year at EPFL in Lausanne, and three years in Louvain-la-Neuve, sharing her/his time between UCLouvain and Keemotion.
Applications should include a detailed resume, including grade sheets for B.Sc. and M.Sc.
Names and complete addresses of referees are welcome.
Please send applications by email to: firstname.lastname@example.org.
|Title||PhD opening in Computer Vision|
|Employer||Université catholique de Louvain|
|Job location||Place de l'Université 1, 1348 Louvain-La-Neuve|
|Published||December 4, 2019|
|Job types||PhD  |
|Fields||Algorithms,   Artificial Intelligence,   Artificial Neural Network,   Computer Communications (Networks),   Programming Languages,   Analysis,   Electrical Engineering,   Computer Vision,   Image Processing,   |