Understanding human activities in everyday and professional scenarios requires fast and accurate 3D hand-object pose estimation. Most existing 3D hand-object pose estimation models rely on dedicated hardware support to address these challenges, limiting their practicality on everyday devices such as smartphones. In this paper, we propose a lightweight solution for hand and object pose estimation for mobile applications, using only RGB images as input, which greatly improves efficiency while maintaining high accuracy, demonstrating the feasibility of real-time hand pose estimation in low-computation environments. This makes a new contribution to interactive applications centered on the human hand under hardware constraints.
History
Thesis type
Thesis (PhD)
Thesis note
Thesis submitted for the Degree of Doctor of Philosophy, Swinburne University of Technology, 2025.