A Survey of Human Pose Recognition Based on WiFi Sensing and Neural Network

With technological advancements and increasing user demands, human action recognition plays a pivotal role in the field of human-computer interaction. Among various sensing devices, WiFi equipment has gained widespread application due to its universal presence. This paper explores the use of WiFi devices and neural network technology to achieve human pose recognition for multiple persons, significantly advancing intelligent environmental perception and human motion analysis. First, we review typical applications of human pose recognition based on computer vision, millimeter-wave radar, and WiFi devices. Second, a human action recognition system is designed using WiFi devices. Subsequently, data preprocessing is performed, innovatively applying phase denoising techniques such as unwrapping and linear transformation to CSI signals, and fusing time-frequency analysis, filtering, and deep learning models to accurately correct phases. Then, leveraging deep learning and attention mechanisms, the system accurately determines the positions of multiple persons and learns human pose. By utilizing prior knowledge of human anatomy and Transformer networks to optimize joint features, action recognition accuracy is enhanced. Analysis of typical systems demonstrates that human action recognition based on WiFi devices and neural networks achieves high precision in pose recognition for multiple persons. Finally, existing challenges and future research directions are discussed.