Motion Capture for Sporting Events Based on Graph Convolutional Neural Networks and Single Target Pose Estimation Algorithms

Human pose estimation refers to accurately estimating the position of human body from a single RGB image and detecting location body. It serves as basis for several computer vision tasks, such tracking, 3D reconstruction, autonomous driving. Improving accuracy has significant implications advancement vision. This paper addresses limitations single-branch networks in estimation. presents top-down single-target approach based on multi-branch self-calibrating combined with graph convolutional neural networks. The study focuses two aspects: detection is athletes appearing sports competitions, followed by estimation, which divided into methods: coordinate regression-based heatmap test-based. To improve test, high-resolution feature map output HRNet used deconvolution recognition.