Estimating the 3D pose of a human from a single image or video is a very challenging computer vision problem since a picture contains little information on the depth. Moreover, humans move in complex ways, and their diversity in appearance and shape is immense. The recent advances in deep learning bring about new levels of accuracy, but they are prone to bias in the training sets that rarely represent the variety present in our global population. Moreover, automatic monitoring raises privacy concerns. I will present personalized solutions that mitigate bias and suggest alternatives to video-based surveillance and its associated dangers.
Helge Rhodin is an Assistant Professor at the University of British Columbia, a member of the computer vision and graphics labs. His research interests range from computer graphics and augmented reality, over 3D computer vision, to machine learning. Helge received the BSc and MSc degree in Computer science from Saarland University. He graduated with a PhD in 2016 for is work at the Max-Planck Institute for Informatics and was a postdoctoral researcher and lecturer at EPFL.