Human Pose Estimation
Overview
Human pose estimation locates body joints and reconstructs 2D or 3D body structure, supporting motion analysis, rehabilitation, sports, animation, and human-robot interaction. Methods range from CNNs and vision transformers that predict keypoints from images to temporal models, graph networks, and probabilistic models that use motion and anatomical priors; radar, WiFi, and wearable sensors offer alternatives when cameras are unsuitable. Benchmarks such as COCO, PoseTrack, Human3.6M, and 3DPW test accuracy across still images, video, and 3D settings. A central challenge is making estimates reliable under occlusion, unusual poses, new environments, and noisy labels while keeping models efficient enough for real-time use.
Latest papers
79 papers13 papers in the last four weeks, up 63% on the four weeks before. 0.1% of all new papers.