Abstract:3D human pose estimation refers to recovering the pose of the human body in the three-dimensional space from the image, which is estimating the coordinates of the key points of the human body in the three-dimensional coordinate system. The human skeleton model composed of these keypoints describes the human pose in the image.The traditional pose estimation method uses motion capture equipment, which requires athletes to wear auxiliary equipment and affects the normal training of athletes. With the development of deep learning, convolutional neural network has shown strong representational capabilities in the process of image feature extraction, and is an effective method to achieve human pose estimation. However, in the actual complex and changeable three-dimensional space, the environment is noisy and the limbs are occluded. The monocular capture of human posture has the problems of low accuracy and instability, and cannot provide sufficient semantic information. To solve this poroblem, this paper designs a multi-view information fusion algorithm. First, a two-dimensional attitude feature extraction module is designed. By connecting and interacting heat maps of different resolving power in parallel, the heat map always maintains a higher resolving power for higher precision. It extracts the two-dimensional human pose from a single perspective, and then uses the multi-view information fusion technology proposed in this paper to complete and correct the pose information from different perspectives to obtain a more accurate two-dimensional pose. Then, through the camera pose transformation, the three-dimensional human pose estimation is realized, and the experimental verification is carried out on the multi-eye machine vision pan-tilt zoom tracking shooting system.

3D Human Pose Estimation Via Spatio-Temporal Matching from Monocular RGB Images.

Marker-Less 3d Human Motion Capture With Monocular Image Sequence And Height-Maps

3D Human Pose Estimation = 2D Pose Estimation + Matching

3D Human Pose Estimation Based on Volumetric Joint Coordinates

STN-enhanced Message Passing Guided by Adversarial Learning for Human Pose Estimation

Three-dimensional Human Pose Estimation Based on Spatio-Temporal Multi-Feature Fusion Network

Towards Locality Similarity Preserving to 3D Human Pose Estimation.

A Survey on Monocular 3D Human Pose Estimation

An Improved 3D Human Pose Estimation Model Based on Temporal Convolution with Gaussian Error Linear Units

3D Human Pose Estimation Based on Multi View Information Fusion

Unified End-to-End YOLOv5-HR-TCM Framework for Automatic 2D/3D Human Pose Estimation for Real-Time Applications

3D Human Pose Estimation in Motion Based on Multi-Stage Regression

3D Human Pose Estimation using Spatio-Temporal Networks with Explicit Occlusion Training

Multi-Person 3d Pose Estimation From Monocular Image Sequences

3D Human Pose Estimation with Spatio-Temporal Criss-Cross Attention

Monocular 3D Human Pose Estimation by Predicting Depth on Joints

3D Human Pose Estimation with 2D Human Pose and Depthmap

3D Human Pose Estimation Via Graph Extended Spatio-Temporal Convolutional Network

An Adaptive Viewpoint Transformation Network for 3D Human Pose Estimation

P-STMO: Pre-Trained Spatial Temporal Many-to-One Model for 3D Human Pose Estimation

Reweighted Sparse Representation with Residual Compensation for 3D Human Pose Estimation from a Single RGB Image