摘要早期动作识别在计算机视觉领域尤其是人机交互方面有广泛的应用，其目的是预测 未完成的动作，与完整动作的事后分类截然不同。本文实现一种基于人体骨骼信息的早 期动作识别算法，我们利用 Kinect 传感器实时准确地捕获人体骨骼的三维数据，据此来 计算归一化距离、生成模板、实现预测。基于此算法我们还实现了一个简单的动作预测 系统，该系统使用 C#语言编写，WPF 搭建用户界面，能同时显示 Kinect 传感器捕获到的 彩色图像和人体骨架，能初步预测 8 个特定的动作。77208

毕业论文关键词 早期动作识别 Kinect 骨骼信息 模板

毕 业 设 计 说 明 书 外 文 摘 要

Title Research and implementation of early action recognition algorithm based on human skeleton information

Abstract Early action recognition has a wide range of applications in computer vision， especially in human-computer interaction applications。The goal is to enable early recognition of unfinished actions as opposed to the after-the-fact classification of completed actions。In this paper,we implement a novel early action recognition algorithm which is based on human skeleton information。We use the Kinect sensor to accurately capture real-time three-dimensional data of the human skeleton,from which to calculate the normalized distance,generate templates and eventually to early recognize actions。In addition,a simple system which is able to early recognize actions has been developed in C# and WPF。This system can simultaneously display color images and human skeleton figure captured by Kinect sensor。At present,this system is capable of early recognizing eight specific actions。

Keywords early action recognition Kinect skeleton information template

本科毕业设计说明书 第 I 页

目 次

1 引言 1

1。1 研究背景 1

1。2 研究现状 1

1。3 本文研究内容及组织结构 2

2 Kinect 及骨骼信息获取 3

2。1 Kinect 深度传感器简介 3

2。2 Kinect 基本接口与使用方法 5

2。3 Kinect 获取骨骼数据 11

3 算法实现 14

3。1 预备知识 14

3。2 算法步骤 16

3。3 关键步骤代码 19

3。4 实验与分析 23

4 系统实现 26

4。1 系统功能及实现代码