摘要
随着科技发展,可穿戴式的传感器研究越发得到重视,显现出低功耗、便携性高、低成本以及使用场景不受限制等优势。其中重要的一方面应用就是人体姿态识别,为了识别日常生活中站姿、跪姿以及卧姿三种不同姿态而进行研究。根据人体姿态识别技术理论分析和应用需求选择单传感器进行姿态识别的方案。选用六轴传感器MPU6050结合STM32单片机硬件的方案,采集三种不同姿态的加速度以及角速度数据,经小波去噪和四元数转换后,基于高斯核函数的一对一支持向量机算法对人体姿态进行分类,模型训练框架基于TensorFlow,验证了利用机器学习算法解决三种人体姿态识别问题的可行性。
With the development of science and technology,more and more attention has been paid to the research of wearable sensors which have the advantages of low power consumption,high portability,low cost,and unlimited use scenarios.One of the most important applications is human posture recognition.Three different postures of standing,kneeling and lying in daily life are studied.According to the human gesture recognition theory and the application requirement,the six-axis sensor MPU6050 and STM32 micro control chip hardware scheme is selected,the acceleration and angular velocity data of three different attitudes are collected.After wavelet denoising and quaternion conversion,a one-to-one support vector machine algorithm based on Gaussian kernel function is selected to classify human gestures.The model training framework is based on TensorFlow,which verifies the feasibility of the use of machine learning algorithms to solve the recognition of three human postures of standing kneeling and lying.
作者
郑毅
宋贺良
王克强
ZHENG Yi;SONG Heliang;WANG Keqiang(North China Research Institute of Electro-Optics,Beijing 100015,China)
出处
《传感技术学报》
CAS
CSCD
北大核心
2023年第3期462-468,共7页
Chinese Journal of Sensors and Actuators
关键词
人体姿态识别
小波去噪
四元数
支持向量机
human posture recognition
wavelet denoising
quaternion
support vector machine
