Speaker: Peter Hardy, University of Southampton
Title: Partial to Whole: The Unsupervised Lifting of Independent 2D Fragments to Complete 3D Poses in Human Pose Estimation
Abstract: 3D Human Pose Estimation has a wide range of applications, from motion capture and early detection of rheumatoid arthritis to modelling human-computer interactions. However, the acquisition of 3D data is both expensive and time-consuming, resulting in the majority of available 3D datasets containing limited actors, performing few actions, and often in controlled environments. Furthermore, the integration of any modern 2D-3D lifting approach with an off-the-shelf 2D detector is currently not feasible due to the incredibly small likelihood of actually acquiring a complete 2D pose at any given point. In light of these challenges, this talk explores the feasibility of unsupervised 3D human pose estimation with the aim of accurately retrieving the 3D pose from partially detected 2D keypoints alone. We then investigate, the potential of ‘filling in the gaps’, where we leverage this partial 3D pose to obtain an accurate, complete 3D skeleton. In this talk, we will cover the complete end-to-end flow of 2D-3D human pose estimation from 2D detection to 3D pose lifting while also introducing the limitations of my approach and where I believe there is potential for future work.
Bio: Peter Hardy is a final year PhD student in Computer Science at the University of Southampton, where he is mentored by Dr Hansung Kim. His research is focused on generative models and 3D reconstructions from single images or videos, with a particular emphasis on modelling 3D human interactions, actions, and gesture recognition. Before his PhD, Peter completed his MSc in Data Science at Lancaster University, during which he undertook an internship with Rinicom Ltd focusing on the detection and elimination of drones in private airspace. Interestingly, Peter’s academic journey began with a BSc in Finance. After two years of working in Amsterdam, he decided to pivot towards a more fulfilling career in computer science.
