Page 62 - 2024F

UEC Int’l Mini-Conference No.53 55

Optimized Graph Neural Network Approach for 3D Human Pose

Estimation

Luis ACEVEDO-BRINGAS , Hiroki TAKAHASHI
∗
Department of Informatics
The University of Electro-Communications
Tokyo, Japan
a2440012@gl.cc.uec.ac.jp
∗

Keywords: 3D Human Pose Estimation, Deep Learning, Graph Convolutional Network (GCN),
Transformers.

Abstract

3D human pose estimation is a fundamental problem in computer vision with applications in
motion analysis, human-computer interaction, augmented reality, and biomechanics. Traditional
methods often rely on handcrafted features or optimization based techniques, which struggle with
occlusions, complex poses, and real-time processing constraints. Recent advancements in deep learn-
ing, particularly graph based approaches, have significantly improved accuracy and robustness in
pose estimation. Graph-based deep learning models, such as Graph Convolutional Networks (GCNs),
effectively capture the spatial and temporal relationships between human body joints. By represent-
ing the human skeleton as a graph, these models leverage local and global dependencies, enabling
more accurate pose predictions, even in challenging scenarios. This research aims to explore effi-
cient graph-based deep learning architectures to enhance the accuracy, computational efficiency, and
generalizability of 3D human pose estimation models.

The author is supported by (SESS) MEXT Scholarship.
*