基于PPG和ConvLSTM的细粒度人体运动识别研究

Research on Fine-Grained Human Motion Recognition Based on PPG and ConvLSTM

摘要: 近年来，光电容积脉搏波描记技术(photoplethysmography, PPG)作为一种无创生理信号监测方法，因其与可穿戴设备的高度兼容性、信号采集准确性以及适用于长期监测的特性，已在人体运动识别与人机交互应用中显示出巨大潜力。本研究通过自制的传感平台，针对15名志愿者执行的三种粗粒度运动(静坐、快走、慢跑)下的五种细粒度运动(上伸臂、下伸臂、抬小臂、弯手指和右摆手)进行了信号采集。创新性地提出了一种基于ConvLSTM网络的运动识别分类算法，并对细粒度运动信号进行分析，揭示了在运动识别应用中，近红外光和绿光信号具有显著的优势。经评估，模型在近红外光和绿光条件下的F1分数分别达到0.951和0.944，展现了高度的识别准确性和可靠性。此外，研究还发现粗粒度运动的强度与信号分类效果之间存在负相关关系，表明运动的复杂度可能会对识别效果产生影响。这些发现不仅证实了PPG信号在精准人体运动识别上的有效性，还为PPG信号应用于可穿戴设备的开发提供了强大技术支持，对促进人体运动科学、人机交互和健康管理领域的创新发展具有重要指导意义。

Abstract: In recent years, photoplethysmography (PPG), as a noninvasive physiological signal monitoring method, has shown great potential in human motion recognition and human-computer interaction applications due to its high compatibility with wearable devices, signal acquisition accuracy, and suitability for long-term monitoring. In this study, signals were acquired by a homemade sensing platform for five fine-grained motions (upward arm extension, downward arm extension, forearm extension, finger bending, and right hand swing) under three types of coarse-grained motions ( sitting, walking, and jogging) executed by 15 volunteers. An innovative motion recognition classification algorithm based on ConvLSTM network was proposed and analyzed for the fine-grained motion signals, revealing that near-infrared light and green light signals have significant advantages in motion recognition applications. The model was evaluated to achieve F1 scores of 0.951 and 0.944 under near-infrared light and green light conditions, respectively, demonstrating high recognition accuracy and reliability. In addition, a negative correlation was found between the intensity of coarse-grained motion and the signal classification effect, suggesting that the complexity of motion may have an impact on the recognition effect. These findings not only confirm the effectiveness of PPG signals in accurate human motion recognition, but also provide strong technical support for the application of PPG signals to the development of wearable devices, which is of great guiding significance in promoting the innovation and development in the fields of human motion science, human-computer interaction and health management.