基于加速度信号的人体前臂动作识别技术研究

发布时间：2018-07-02 12:06

本文选题：加速度 + 前臂动作识别　；参考：《哈尔滨工程大学》2016年硕士论文

【摘要】：作为一种新兴的自然人机交互方式,基于加速度信号的人体前臂动作识别技术在智能人机交互、体育锻炼、体感游戏、虚拟现实、智能控制等领域得到了广泛的应用,是广大研究人员研究的热点。虽然近年来发展迅速,但是该研究在加速度信号数据采集和处理方案以及高精度的动作识别算法实现方面还面临着诸多技术难点。本课题围绕上述问题展开了一系列研究工作。首先本文详细阐述了人体动作识别技术的研究背景及意义,并对国内外研究现状和前臂动作识别技术的主要研究方法进行了介绍,然后给出了前臂动作识别研究的系统方案。其中,包括了对加速度数据的硬件采集平台的选取以及人体前臂动作识别算法的设计。接着对于本文所设计的人体前臂动作识别算法,分别从数据预处理、特征选择与提取以及分类器设计等方面对算法的实现进行分析与设计。结合小波分析相关理论,选取小波阈值去噪方法对原始加速度信号进行去噪处理,并对去噪后的数据进行归一化,然后给出一种基于小波包分解的三维全信息能量特征参数构造方法,同时从时域和时频特征中选择特征参数来构建特征向量。在分类器设计方面,选择了 SVM分类方法,并以LIBSVM和MATLAB作分类算法实现的软件平台。最后通过设计具体的验证实验对所提出的前臂动作识别系统方案进行验证。确定采集装置安放位置和自定义了七种简单前臂动作类型,通过采集10位实验对象七种前臂动作加速度信息构成整个实验原始数据集。对每组加速度信号序列进行预处理操作,并构造特征向量来建立性能较好的SVM分类器。通过设计实验进行对比,采取交叉验证的方法选用核函数,采用网格搜索算法求解最优核参数对(C,g),建立了最优SVM分类判决模型。最终在“指定数据集”和“非指定数据集”上对SVM分类算法进行验证分析,对实验结果进行分析,验证了模型的性能和研究方案的可行性。
[Abstract]:As a new natural human-computer interaction technology, the technology of human forearm action recognition based on acceleration signal has been widely used in the fields of intelligent human-computer interaction, physical exercise, physical training, body feeling games, virtual reality, intelligent control and so on. It is a hot topic for researchers. Although the research has been developing rapidly in recent years, the research still faces many technical difficulties in the data acquisition and processing of acceleration signal and the realization of high-precision motion recognition algorithm. This topic has carried out a series of research work around the above problems. In this paper, the research background and significance of human motion recognition technology are described in detail, and the present research situation and the main research methods of forearm motion recognition technology are introduced, and then the system scheme of forearm motion recognition research is presented. Among them, it includes the selection of hardware acquisition platform for acceleration data and the design of human forearm motion recognition algorithm. Then, the realization of the algorithm is analyzed and designed from the aspects of data preprocessing, feature selection and extraction, classifier design and so on. Combined with the theory of wavelet analysis, the wavelet threshold denoising method is selected to Denoise the original acceleration signal, and the de-noised data is normalized. Then a method based on wavelet packet decomposition is presented to construct the three dimensional full information energy feature parameters, and the feature parameters are selected from the time domain and time frequency features to construct the feature vector. In the design of classifier, SVM classification method is selected, and the software platform of classification algorithm is based on LIBSVM and MATLAB. Finally, the scheme of forearm motion recognition system is verified by designing specific verification experiment. The location of the acquisition device is determined and seven types of simple forearm actions are defined. The original data set of the whole experiment is constructed by collecting seven kinds of acceleration information of the forearm action of the 10-bit experimental object. Every group of acceleration signal sequence is preprocessed and eigenvector is constructed to build SVM classifier with better performance. Through the comparison of design experiments, the kernel function is selected by cross-validation method, and the optimal kernel parameter pair (Cogg) is solved by using the grid search algorithm, and the optimal SVM classification decision model is established. Finally, the SVM classification algorithm is verified and analyzed on "specified data set" and "non-specified data set", and the experimental results are analyzed to verify the performance of the model and the feasibility of the research scheme.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】