A Survey of Deep Learning-Based Human Pose Estimation: Methods, Datasets, and Evaluation Metrics

Human pose estimation has emerged as one of the most prominent research directions in computer vision in recent years. This technology aims to acquire human pose information from images or videos, demonstrating immense potential in enhancing human life. With the rise of deep learning, methods based on deep neural networks have been widely adopted, replacing traditional feature-based or model-based approaches to achieve more effective estimation of human pose information. This paper systematically reviews recent advances in the field, categorizing relevant research methodologies into four classes: approaches based on Convolutional Neural Networks (CNN), Generative Adversarial Networks (GAN), Graph Convolutional Networks (GCN), and Transformer architectures. Representative works under each category are introduced. The article also provides detailed descriptions of commonly used 2D and 3D pose estimation datasets. Finally, it presents current performance evaluation metrics for both 3D and 2D pose estimation, offering researchers a comprehensive and in-depth understanding of the field.