Human-Centric Multimodal AI for Motion Understanding and Reasoning

The School of EECS is hosting the following HDR Progress Review 1 Confirmation Seminar: 

 

Human-Centric Multimodal AI for Motion Understanding and Reasoning

 

Speaker: Houke Zhao

 

Abstract: Human-centric AI has become increasingly important for healthcare, rehabilitation, sports analysis, assistive technology, and embodied intelligence. However, many existing systems are still built on assumptions that limit their use in real-world and inclusive settings. Human pose estimation often requires raw images or videos, which may expose sensitive identity, appearance, body shape, gait, health, and environmental information. Meanwhile, 3D human modeling and multimodal motion understanding are still largely designed around able-bodied people, making them less reliable for amputees, disabled athletes, and assistive-device users.

 

This confirmation presents my PhD research on privacy-preserving and accessible human motion understanding. The completed work investigates frequency-domain privacy protection for 2D human pose estimation, amputee-specific 3D human mesh recovery, and multimodal reasoning for assistive human motion. The future work will further explore dataset and benchmark construction for disabled motion understanding, as well as scene-aware disabled human motion modeling. Overall, this research aims to develop human-centric AI systems that are not only accurate, but also privacy-aware, inclusive, and suitable for real-world deployment.

 

Bio: Houke Zhao is a second-year Ph.D student at School of EECS, University of Queensland, under the supervision of AsPr Xin Yu. His research centres on human-centric understanding and reasoning.