2D Human pose estimation: a survey

Human pose estimation aims at localizing human anatomical keypoints or body parts in the input data (e.g., images, videos, signals). It forms a crucial component enabling machines to have an insightful understanding of behaviors humans, and has become salient problem computer vision related fields. Deep learning techniques allow feature representations directly from data, significantly pushing performance boundary estimation. In this paper, we reap recent achievements 2D methods present comprehensive survey. Briefly, existing approaches put their efforts three directions, namely network architecture design, training refinement, post processing. Network design looks models, extracting more robust features for keypoint recognition localization. refinement tap into neural networks improve representational ability models. Post processing further incorporates model-agnostic polishing strategies detection. More than 200 research contributions are involved survey, covering methodological frameworks, common benchmark datasets, evaluation metrics, comparisons. We seek provide researchers with systematic review on estimation, allowing them acquire grand panorama better identify future directions.