基于惯性传感器的运动识别系统_0533..doc

摘要 基于惯性传感器的运动识别系统是模式识别的一个新兴领域，克服了传统基于视频的动作识别的诸多缺点和限制，具有更高的可操作性和实用性。所以本文首先着重介绍了如何利用动作的进行识别的算法中，比较常用的算法有、支持向量机模型不是很直观，而且相对来说比较复杂，因此s均值聚类算法，本文的工作流程如下：首先实验对象手中数据自动截取最后利用选定的算法识别，用于识别的几种动作为上、、向左、向右和画圈，对动作进行特征提取，最后采用-means均值聚类进行识别，识别率-means均值聚类, FFT  ABSTRACT Inertial sensor based motion recognition system is an emerging field. It overcomes the disadvantage of traditional video based gesture recognition, and has higher maneuverability and practicability. So this article introduces how to classify motions with inertial sensor, and its working principle. The most common algorithm for motion recognition is Neural Networks (NN), support vector machine (SVM), hidden Markov models (HMM), etc.. These algorithms are not so intuitive and relatively complex in algorithm. So this paper adopts k-means clustering for inertial sensor based motion recognition. The workflow is as flows: first of all, the inertial sensor fixed in subjects’ hand is used to collect motion information; and then, the data is transmitted to PC by wireless transmission module, followed by data preprocessing, feature extraction and selection. Finally, the motion is recognized by k-means clustering，This paper exploits the inertial sensors for the recognition of the following motions: up, down, left, right and the circle. The algorithm used for feature selection is Fast Fourier Transform (FFT), and the algorithm for recognition is K-means clustering. So this article is focused on feature extraction, and motion recognition. The corresponding algorithms are fast Fourier transform, and K-means clustering algorithm. KEY WORDS：Inertial sensors, K-means clustering, FFT  第一章 前言 4 1.1研究的背景和意义 4 1.2基于惯性传感器的运动识别国内外现状 5 1.3传感器的分类 6 1.4本文的组织结构 10 第二章 数据采集和特征提取算法 12 2.1 惯性传感器的结构 12 2.2 数据采集 12 2.3 数据的特征提取 13 2.4 离散傅里叶变换(DFT) 16 2.4.1 快速傅里叶变换(FFT) 17 2.4.2快速傅里叶变换FFT 的应用 18 2.4.3 算法基本原理 19 第三章 基于惯性传感器的运动识别 21 3.1 运动识别主要算法 21 3.2K-means算法介绍 22 3.3K－means 文本聚类算法优化 23 3.4 初始聚类中心的选择 24 3.5K一均值聚类算法的算法流程 24 3.6 SOM算法介绍