摘要：基于Kinect体感摄像头依次实现了实景现场三维重构和二维地图生成，并基于地图实现了障碍条件下的移动机器人路径规划与仿真。首先利用OpenNI框架驱动Kinect获取某个场景多角度下的点云数据，然后对点云进行去除无效点、下采样等滤波操作；接着对点云数据进行ICP配准，使多幅点云通过矩阵变换转换到同一坐标系下；对场景点云进行平面提取、障碍物聚类分割、障碍物拟合等工作；随后对处理后的点云进行重采样、平滑等操作以消除误差和点云冗余；然后对生成的场景进行贪婪三角化曲面重建和泊松重建，并拼接场景。最后将三维地图投影到二维平面，在Matlab平台实现了基于人工势场法的移动机器人的路径规划。实验表明基于Kinect进行3D视觉导航是一种可行的低成本策略。
目录
摘要 １
关键词 １
Abstract １
引言
1 选题背景 ２
1.1 问题的提出 ２
1.2国内外研究状况 ２
1.2.1国外研究状况 ２
1.2.2国内研究状况 ２
2 开发平台及应用技术 ２
2.1 Kinect ２
2.2 OpenNI ３
2.3 Microsoft Visual Studio 2010 ３
2.4 MATLAB ３
3 实验流程 ４
4 数据获取及处理 ４
4.1 数据采样 ４
4.1.1 点云 ４
4.1.2 获取点云 ５
4.2 数据预处理 ５
4.2.1去除无效点 ５
4.2.2 基于体素化栅格的下采样滤波 ６
4.2.2.1 简介 ６
4.2.2.2 算法分析 ６
5 配准还原 ８
5.1 简介 ８
5.2 总体流程 ９
5.3 参数和结果分析 ９
6 分割 １１
6.1 平面分割 １１
6.1.1 随机采样一致性算法(RANSAC) １１
6.1.2 结果分析 １２
6.2 障碍物分割 １４
6
 *51今日免费论文网|www.jxszl.com +Q: %3^5`1^9`1^6^0`7^2# 
.2.1 KdTree １４
6.2.1.1 简介 １４
6.2.1.2 构造算法 １５
6.2.1.3 构造实例 １５
6.2.2 欧氏聚类分割 １７
6.2.2.1 简介 １７
6.2.2.2 实验结果 １７
7 障碍物投影 １７
7.1 原理与实验过程 １８
7.2 实验结果 １８
8 障碍物拟合 １９
8.1 最小二乘法拟合 １９
8.1.1 拟合思路 １９
8.1.2 计算过程 １９
8.1.3 结果展示 ２１
9 曲面重建 ２２
9.1 法线估计 ２２
9.1.1 简介 ２２
9.1.2 实验结果 ２３
9.2 平滑 ２３
9.3 泊松重建 ２４
9.3.1 算法原理 ２４
9.3.2 实验结果 ２４
9.4 贪婪三角化 ２５
10 基于人工势场法的移动机器人路径规划 ２６
10.1人工势场法简介 ２６
10.2 算法分析 ２６
10.3 结果分析与演示 ２８
10.3.1情景1 ２８
10.3.2 情景2 ３１
11 不足与展望 ３３
11.1 不足 ３３
11.2展望 ３４
致谢 ３４
参考文献 ３４
附录A ３４
基于Kinect传感器的场景三维重建
3D Reconstruction of Reality Scene Based on Kinect Sensor
Student majoring in network engineering LIN Xin
Tutor XIE Yuancheng
Abstract：Threedimensional reconstruction of realistic scene is conducted based on Kinect somatosensory camera and a twodimensional map is generated according to that. Meanwhile, the path planning and simulation of mobile robots under the obstacles based on the map is realized. Firstly, OpenNI frame is used to drive the Kinect in order to obtain the multi frame point cloud data from several degree angles of view. And then the invalid points are removed from cloud points under the sampling and other filtering operation. Point clouds are used for ICP registration, making the multiple point clouds to be transformed to the same coordinate system through the matrix transformation in the next step. The subsequent work is to carry out plane extraction from the original point clouds, obstacle clustering segmentation, obstacle fitting, etc. After that, resampling, smoothing and other operations of point clouds are done to eliminate errors and point cloud redundancy. Later on, the generated scene is reconstructed by greedy triangulation and poisson reconstruction. Finally, the 3D map is projected onto the 2D plane and the path planning of mobile robots based on the artificial potential field method is realized on the Matlab platform. The experiment indicates that the implement of 3D visual navigation based on Kinect is a feasible and lowcost strategy.

原文链接：http://www.jxszl.com/jsj/wljs/46049.html