摘要自从微软kinect深度摄像机出现，由于kinect深度摄像机拍摄的深度图分辨率高而且成本低，这短短几年时间内依据深度信息的人机交互领域研究，特别是体感游戏上的技术有了突飞猛进的发展。本文将先叙述对传统的基于彩色图像的姿态检测系统，包括其系统的检测流程，检测原理以及检测测试结果。然后引入了深度信息，对深度图进行区域生长法阈值分割得到三元图（原始图像被分成前景区域，不确定区域以及背景区域）以屏蔽掉复杂无关的背景，减少非目标区域对前景的干扰影响。以此为基础，对提取前景的Grab cut算法进行了改进和扩充，并构建了基于彩色图与Kinect深度图像的姿态检测系统。65594

毕业论文关键词  姿态检测；Kinect深度图；三元图；前景提取；区域生长法；Grab cut算法.

毕业设计说明书（论文）外文摘要

Title    Pose detection method based on the Microsoft                     

                Kinect somatosensory game controller                                                

Abstract

Since the Microsoft Kinect depth of the camera appears, the depth of the kinect depth camera shot high resolution and low cost, which just a few years research in the field of human-computer interaction based on depth information, especially on the somatosensory game technology has been rapid development.This article will first describe the traditional attitude detection system based on color images, including the detection process of the system, the detection principle and the detection test results. Then the introduction of in-depth information on the depth chart the region growing threshold segmentation ternary diagram (the original image is pided into a foreground region, uncertain area and the background area) masked complex has nothing to do with the background, reducing the interference of non-target areas on the foreground affected. On this basis, the extraction prospects Grab cut algorithm has been improved and expanded and built color diagram with the Kinect depth image-based gesture detection system.

Keywords  pose detection; Kinect depth map; Trimap; foreground extraction; region growing method; Grab cut algorithm.

目   次

1 绪论 3

1.1 研究内容 3

1.2 研究意义 3

1.3 研究现状与发展趋势 3

2 基于彩色图像的人体部位识别 4

2.1 三元图（TRIMAP） 5

2.2 GRABCUT算法 6

2.2.1 硬分割 6

2.2.2 边界更新并目标前景恢复 8

2.2.3 混合高斯模型的缺点 9

2.3 图像解析 9

2.3.1 边缘图像 10

2.3.2  边缘模型 10

2.3.3  区域模型 10

2.4 外观模型的优化 11

2.4.1  位置先验 11

2.5  分析新图片的外观模型和前景分割 11

2.5.1  颜色模型 12

3 对于2D彩色静止图像的姿态识别测试