Real-time 3D human pose estimation system based on multi-views

Multi-view-based real-time 3D human pose estimation systems have seen widespread application across various domains, including sports, virtual reality, film production, and video surveillance. Despite significant advancements at the algorithmic level, the challenge of real-time performance remains a...

Full description

Saved in:
Bibliographic Details
Published in:Journal of physics. Conference series 2024-11, Vol.2897 (1), p.012036
Main Authors: Cai, Huiqi, Han, Peng
Format: Article
Language:English
Subjects:
Parallel processing
Pose estimation
Real time
Synchronism
Time synchronization
Virtual reality
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Multi-view-based real-time 3D human pose estimation systems have seen widespread application across various domains, including sports, virtual reality, film production, and video surveillance. Despite significant advancements at the algorithmic level, the challenge of real-time performance remains a critical factor that restricts the full application potential of these systems. The synchronization and fusion of data from multiple cameras, high computational complexity, and the rigidity of system architecture contribute to the difficulty in balancing speed and accuracy within existing systems. In response to these challenges, a real-time 3D human pose estimation system based on multiple views has been proposed. This system achieves synchronization of data from multi-views, incorporates a lightweight model design, and implements parallel processing within its architecture. As a result, it is capable of outputting high-precision 3D human poses at a real-time speed of 25.56 FPS.
ISSN:1742-6588
1742-6596
DOI:10.1088/1742-6596/2897/1/012036