A Systematic Review of Recent Deep Learning Approaches for 3D Human Pose Estimation.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Deep learning has revolutionized 3D human pose estimation.
• Existing surveys often lack depth or novel categorization.
• A comprehensive, up-to-date review is needed for monocular, video, and multi-view approaches.

Purpose of the Study:

• To systematically review recent deep learning methods for 3D human pose estimation.
• To provide a novel perspective by categorizing methods beyond learning paradigms.
• To serve as a key resource for researchers in the field.

Main Methods:

• Systematic literature review methodology.
• Focus on monocular images, videos, and multi-view cameras.
• Categorization based on inter-frame models for videos and relative/absolute poses for multi-person estimation.

Main Results:

• Comparison of image-based approaches, including 2D models.
• Analysis of video-based methods based on temporal modeling.
• Differentiation between relative and absolute pose estimation in multi-person scenarios.

Conclusions:

• The survey provides a meticulous overview of state-of-the-art 3D human pose estimation.
• Novel categorization enhances understanding of diverse methodologies.
• Identifies promising future research directions in deep learning for pose estimation.