Computer Vision & Deep Learning based Realtime and Pre-Recorded Human Pose Estimation

This research study utilizes computer vision to estimate multi-person human pose from real-time and pre-recorded video. Computer vision examines human posture detection from RGB images. The proposed method works well for gesture control, gaming, human tracking, action detection, and action tracking. Tracking semantic important points is pose estimation. Two-dimensional human pose estimation predicts the spatial placement of key human body points from images and videos. Several anatomical areas use hand-crafted feature extraction methods to estimate two-dimensional human position. Visual input data and human body component locations are used to estimate human pose. OpenCV and Mediapipe detected 33 posture landmarks in our research. Estimating human body state requires modeling. Model-based methods are used to describe and infer human posture in 2D or 3D. This research uses the BlazePose GHUM 3D Pose Landmark Model for 2D human pose estimation. Output poses include x, y, z, and visibility. Image width and height are X and Y landmark coordinates. landmark depth Z Image shows landmark. Image or video resolution, the number of persons in the scene, and anomalies might impact visibility point accuracy, which was 0.969%.