Center for Research in Comptuer Vision
Center for Research in Comptuer Vision

Seminar Announcement

Rate-invariant analysis of trajectories on Kendall shape space-applications to 3D action recognition and physical performance assessment

Dr. Boulbaba Ben Amor of Mines-Telecom Institute (IMT) Lille Douai, France

Thursday, July 6, 2017 · 1:30PM · HEC 119

The analysis of imaged shapes, either static 2D or 3D, is a well-explored research topic with several elegant literature contributions, such as invariance to rigid transformations, robustness to non-rigid (and even elastic) deformations, tolerance to missing data of shape representations and related applications in recognition, retrieval, clustering, prediction and classification. Nevertheless, what about analyzing dynamic shapes (i.e. the shape's animation over time)? Is there any natural extension from (static) shape analysis? What are the new challenges when introducing the temporal dimension? Which target applications in computer vision and medical diagnosis?

In this talk, we present a novel geometric methodology in modeling dynamic skeletal shapes. Three-dimensional skeletons are naturally mapped into Kendall's shape space to build time-parametrized trajectories. To allow accurate inference and statistical analysis of underlying realizations (i.e. 3D sequences) and derived feature functional data, a rate-invariant metric was rigorously defined. We introduce a comprehensive toolbox for comparing, aligning, filtering, smoothing and computing statistical summaries of shape trajectories was developed. We describe as well a novel set of space-time features to quantify spatial and temporal symmetry, the balance and the shape's velocity. An experimental illustration of our geometric framework will be shown on (1) 3D action recognition, an active research topic in computer vision, and (2) Human body's kinematic analysis for physical performance assessment, a promising research field with several applications in healthcare, athletic training, culture, etc.

Boulbaba Ben Amor is full professor with the Mines-Telecom Institute (IMT) Lille Douai in France and member of the UMR CNRS Research Center CRIStAL. He is currently visiting professor at the SSAMG (Statistical Shape Analysis and Modeling Group) in Florida State University under a Fulbright research scholarship (2016-2017). In 2014, he also visited the same group for 8 months. He was the general chair of the 6th edition of the international workshop RFMI (Representation, analysis and recognition of shape and motion FroM Imaging data), endorsed by the IAPR and published by Springer. He holds a "Habilitation to supervise research" from University of Lille (France), since 2014. He earned a Ph.D. in Computer Science from the Ecole Centrale de Lyon, in 2006. He is the founding director of the Master program "Cybersecurity Engineering" at IMT Lille Douai. His research topics lie to 3D shape and their dynamics analysis for advanced human behavior analysis and assessment. He published 11 journals papers, 4 book chapters, and more than 50 conference papers on these topics.