Konstantinos Papoutsakis - Computer vision methods for human motion capture and activity
recognition

September, 08 at 11:00 (MSC)

Abstract:

Computer vision is an area of artificial intelligence aimed at developing the perceptual capabilities to enable computers to “see” and understand their environment through acquisition, processing, and analysis of image and video sensory data. In this talk, we provide an overview of our work related to computer vision-based methods for the visual perception of human motion in 3D and for the semantic interpretation of human-centered actions in videos using advanced machine learning and deep learning-based techniques. We mainly focus on estimating the articulated pose and tracking the motion of the human hands and the full body in 3D, possibly while interacting with objects, on modeling the spatio-temporal dynamics of the observed entities in videos towards efficient temporal co-segmentation and classification of human actions of varying complexity in videos (from hand gestures to complex human-object interactions). Finally, we provide examples of how our work can support the development of vision systems for applications in the fields of robotics, human-robot interaction in the context of smart environments and assistive robotics, human-robot collaboration, and human-centered monitoring of activities during manufacturing tasks in industrial environments.