基于HMMs和SVM的人体日常动作序列分割识别研究

Research on segmentation and recognition of human daily action sequence based on HMMs and SVM

随着微机电系统(MEMS)研究的精细化,人体传感器网络(简称体感网)技术在医疗监护领域有了长足发展,而人体动作分析与识别是体感网中富有挑战性的研究课题.采用动态隐马尔可夫模型(HMMs)方法对基于用体感网技术的人体动作序列进行了分割,并且对分割精准度进行了度量分析.从实验结果可以看到,动态HMMs方法优于LIR和Top-Down方法,其分割精准度达到了80%以上.对分割后的数据提取均值、方差等特征,采用支持向量机(SVM)方法分类识别的结果表明所提分割方法具有良好的稳健性,平均识别准确率在89%左右,与手动分割接近.

With the refinement of the study of the micro-electro-mechanical system （MEMS）,the application of body sensor networks （BSN）has developed rapidly in the field of medical care.Human motion analysis and recognition are challenging research topics in the BSN.An approach of the dynamic hidden Markov models （HMMs）is proposed to segment the time series of the activities based on BSN.A method of the precision measurement is used to test the approach of the segmentation. The experimental results show that the proposed approach is prior to the LIR and Top-Down methods and the segmentation precision of the dynamic HMMs is above 80%.The features of the data obtained from segmentation,such as mean,variance,etc.are extracted.The results of the recognition by support vector machine （SVM）show the robustness of the proposed segmentation method.The mean recognition accuracy is about 89%,which is near the manual segmentation.