A Review of 2D Human Pose Estimation Based on Convolutional Neural Networks

The rapid progress in deep learning has positioned 2D human pose estimation as a critical foundationfor numerous computer vision applications. The efficiency and precision of these methods are vital for theirpractical implementation. This paper presents a comprehensive review of contemporary 2D human poseestimation techniques utilizing convolutional neural networks (CNNs), organized into two main categories:human detection combined with joint regression fusion and joint detection with clustering algorithms.Additionally, it provides an overview of widely used datasets and evaluation standards. The study furtherexamines existing challenges and explores potential future directions in the field, offering valuable insightsto guide subsequent research efforts.