Human Centric Visual Analysis - Hand, Gesture, Pose, Action, and Beyond

Human centric visual analysis tasks are essential to computer vision since humans are the key element for cameras to analyze. In this talk, I will mainly focus on 4 visual analysis tasks on human hand, gesture, pose, and action respectively. More specifically, I will introduce the following 4 tasks: 1) hand gesture recognition using RGBD camera, 2) 3D hand shape and pose estimation from a single RGB image, 3) human pose estimation and tracking by modeling temporal dynamics, and 4) weakly-guided self-supervised pretraining for action detection, which I have published in IEEE Transactions on Multimedia 2013, CVPR 2019, CVPR 2021, and AAAI 2023.

This webinar covers human centric visual analysis using various input modalities including RGBD, RGB image and video. Different state-of-the-art modeling techniques will be introduced, such as Finger Earth Mover's Distance, Graph Neural Networks, and Self-Supervised Learning, whose effectiveness are demonstrated in those tasks. With finer and better human analysis from hand to gesture, pose, and action, at the end I will briefly discuss how to extend to other tasks beyond, such as ReID, tracking, and Human-Computer-Interaction, etc.