Person Re-identification in Cross-Domain Adaptive Network

Video analytics is an active fields of research, where state-of-the-art systems rely on a variety of computer vision and machine learning techniques for accurate modeling and recognition from large-scale video datasets. Person re-identification is a key problem found in numerous application areas, e.g., video surveillance, summarization, and sports analysis, and seeks to match people across non-overlapping views in a multi-camera system. However, this remains a challenging problem because the appearance of individuals varies considerably across cameras viewpoints (pose, illumination, etc.), and due to the non-rigid structure of individuals. This project will focus on developing accurate visual recognition models that allow for person re-identification in sports video analytics application of interest for SPORTLOGiQ Inc., leading to person tracking, activity recognition and group behavior understanding over a distributed network of cameras. Designing accurate recognition systems for these applications typically gives rise to several challenges because it involves learning complex models using large weakly-annotated data sets that incorporate domain shifts, subtle noise, variations and uncertainties embedded in real-world signals.TO BE CONT’D

Faculty Supervisor:

Éric Granger

Student:

Partner:

Sportlogiq

Discipline:

Computer science

Sector:

Information and Communications Technology; Technology

University:

École de technologie supérieure

Program: