Pose Estimation
تقدير الوضعية
Also known as: Keypoint Detection, Skeleton Detection
Pose estimation locates anatomical keypoints (shoulders, elbows, knees, ankles) on every person in a frame, producing a stick-figure skeleton. Top-down approaches (RTMPose, ViTPose) first detect bounding boxes then regress keypoints; bottom-up approaches (OpenPose, HigherHRNet) detect all keypoints first then group. FI Tech uses pose for fall detection (sudden Y-coordinate drop…
Definition
Pose estimation locates anatomical keypoints (shoulders, elbows, knees, ankles) on every person in a frame, producing a stick-figure skeleton. Top-down approaches (RTMPose, ViTPose) first detect bounding boxes then regress keypoints; bottom-up approaches (OpenPose, HigherHRNet) detect all keypoints first then group. FI Tech uses pose for fall detection (sudden Y-coordinate drop of head and torso), ergonomic monitoring (sustained back angles above 45° in lifting tasks), and unsafe-act detection like reaching into machine guards. Sapiens (Meta, 2024) and RTMPose-X are the current accuracy leaders for industrial settings.
