人体坐姿识别智能终端的压力传感器排布问题研究  

Study on pressure sensors arrangement problem of human sittingposture recognition intelligent terminal

作　　者：杜英魁[1] 姚俊豪 王馨鹤 刘洪安[2] 李若溪[2] 原忠虎[1] DU Yingkui;YAO Junhao;WANG Xinhe;LIU Hongan;LI Ruoxi;YUAN Zhonghu(College of Information Engineering,Shenyang University,Shenyang 110044,China;The Fourth People’s Hospital of Shenyang,Shenyang 110031,China)

机构地区：[1]沈阳大学信息工程学院,辽宁沈阳110044 [2]沈阳市第四人民医院,辽宁沈阳110031

出　　处：《传感器与微系统》2020年第7期54-57,共4页Transducer and Microsystem Technologies

基　　金：辽宁省重点研发计划资助项目(2019JH2/10300014);辽宁省重点研发计划指导计划资助项目(2018104013);沈阳市“双百工程”重大科技成果转化项目(Z19-1-002)。

摘　　要：在基于电阻式薄膜压力传感器的人体坐姿识别研究中,传感器排布方式是一个影响识别准确率和硬件成本的基础性问题。提出了一种基于时间序列欧氏距离相似性度量的传感器排布选择算法,定义了不同坐姿下的归一化压力时间序列数据,根据不同传感器之间时间序列数据欧氏距离相似度的差异显著性,对每个感应区域的压力传感器进行排布选择优化。利用随机森林和叠层支持向量机(SVM)分类模型进行了十折交叉对比实验,实验结果表明:该方法可减少25%的传感器数量并提升人体坐姿识别准确率,验证了所提方法的有效性。In human sitting position recognition research based on resistive thin film pressure sensor,the arrangement of pressure sensors is a key problem,which affects the recognition accuracy and hardware cost.A sensor arrangement and selection algorithm is proposed based on Euclidean distance similarity measurement of time series.The normalized pressure time series data of different postures is defined.In the cross-contrast experiments,the random forest and stack support vector machine are utilized.Expenmental results show that 25 percent sensors are reduced but the accuracy of recognition is improved.The effectiveness of the algorithm is validated.

关 键 词：坐姿识别 压力传感器 时间序列 随机森林 支持向量机 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置]