Abstract: To further improve the accuracy rate of daily action recognition system for the aged, an action recognition scheme for the aged based on flexible sensor is presented in this thesis. Through the combination of flexible sensor with three dimensional gravity acceleration sensor, the action sequence data collection, processing and recognition method for the aged has been formed. According to the waveform characters of human action sequence, it substitutes the automatic cutting algorithm based on Fourier transform for traditional manual intervened data preprocessing and feature extraction mode; by utilizing Stacking ensemble learning technology, it takes random forest and Naive Bayes as the base classifier and takes logical regression algorithm as secondary classifier to form the classification model. Compared with former traditional technology depending on single three dimensional gravity acceleration sensor and single strong classification model for data collection and analysis, the accuracy rate of recognition of the method proposed in this thesis has been improved significantly and the method has been tested in clinical application. The experimental results show that the action recognition technology for the aged based on flexible sensor and ensemble learning can reach the accuracy rate of above 90% at the time of recognizing various types of action simultaneously.

Key words: flexible sensor, waveform processing, the aged, action analysis, ensemble learning

摘要： 为进一步提高老年人日常动作识别系统的准确率，提出一种基于柔性传感器的老年人动作识别方案。通过柔性传感器与三维重力加速度传感器相结合，形成了老年人动作时序数据采集、处理和识别方法，根据人体动作时序波形特征，采用基于快速傅里叶变换的自动切割算法替代传统人工干预的数据预处理和特征提取方式；利用Stacking集成学习技术，将随机森林和朴素贝叶斯作为基分类器，以逻辑回归算法作为次级分类器生成分类模型。与以往依赖单一三维重力加速度传感器以及单一强分类模型进行数据采集和分析的传统技术相比，提出的方法在识别准确率有显著提升，并在临床应用中得到检验。实验结果表明，基于柔性传感器和集成学习的老年人动作识别技术在同时识别多种类型动作时能达到90%以上的准确率。

关键词: 柔性传感器, 波形处理, 老年人, 步态分析, 集成学习