Human Motion Pose Recognition System Based on Convolutional Neural Networks

This paper proposes a human motion pose recognition system based on convolutional neural networks (CNNs). The system consists of four modules: data acquisition, preprocessing, feature extraction, and pose recognition, capable of extracting high-dimensional features from various input forms (RGB, depth, skeletal data). The feature extraction module employs a novel network architecture, while the pose estimation module includes sub-networks for keypoint regression and pose classification. Experimental results demonstrate superior performance of the system in both keypoint regression and pose classification tasks on public datasets, highlighting its outstanding pose estimation and understanding capabilities.