基于RGB-D的SLAM定位与GPU三维建图方法研究1

专 业 学 位 硕 士 学 位 论 文

基于 RGB-D 的 SLAM 定位与 GPU 三维建图

方法研究

作 者 姓 名： 王 振 龙

工 程 领 域： 软 件 工 程

学 号： 31617067

指 导 教 师： 周宽久 教授

完 成 日 期： 2018 年 3 月 25 日

- I -

而本文正是在此基础上提出了一个采用 RBG-D 图像序列的同步定位与三维建图方

文可分为基于特征的视觉里程计设计和拓扑图优化的 GPU 三维建图。其一，对相邻的

两张 RGB-D图像使用 ORB特征提取算法进行特征提取，特征匹配算法使用的是 FLAAN，

可以进行特征匹配，使用基于距离筛选的策略和 RANSAC 算法进行特征点匹配对的筛

选优化，最后使用 PNP 计算相机位姿。当输入 RGB-D 图像序列时，则可以实时输出相

向来筛选关键帧减小点云噪声。

最后使用 GPU 加速点云间坐标转换，然后将点云图转换为八叉树地图用以存储以

Research on SLAM Location and GPU 3D Mapping Method Based on

RGB-D

Abstract

The traditional sensor's SLAM usually needs to place identification marks in the

environment. The vision based SLAM system can realize the requirement of positioning and

mapping without changing the environment, and is more flexible and adaptable to the complex

environment. But the visual based SLAM system will consume a lot of computing power when

dealing with the image information of the sensor and modeling the 3D modeling. It is difficult

matching algorithm is FLAAN, which can match the feature. The distance selection strategy

and RANSAC algorithm are used to optimize the matching of feature points. Finally, PNP is

used to calculate the position and posture of the camera. When the RGB-D image sequence is

input, the position and posture of the camera can be output in real time. Secondly, according to

the position and posture of the camera in the previous part, the motion information of the camera

is stored by the topology graph, and the cumulative error generated in the camera movement is

reduced by the closed loop detection optimization. At the same time, the camera moving

distance and the camera direction are used to screen the key frame to reduce the cloud noise.

Finally, we use GPU to accelerate point cloud coordinate transformation, and then convert

point cloud to octree map to store location requirements. Octree map is a 3D map based on

- III -

摘 要 ............................................................................................................................. I

Abstract ............................................................................................................................. II

1 绪论 .............................................................................................................................. 1

1.1 研究背景及意义 ............................................................................................... 1

1.2 SLAM 问题概述 ............................................................................................... 2

1.3 基于 RGB-D 相机的 SLAM 方法研究现状 ................................................... 2

1.4 本文主要工作 ................................................................................................... 4

2 SLAM 系统的相关技术综述 ...................................................................................... 6

2.1 相机成像模型 ................................................................................................... 6

2.1.1 针孔成像模型 ........................................................................................ 6

2.1.2 坐标系变换关系 .................................................................................... 6

2.4 三维地图 ......................................................................................................... 16

2.4.1 点云图 .................................................................................................. 16

2.4.2 八叉树地图 .......................................................................................... 17

3 基于图像特征的视觉 SLAM 设计 ........................................................................... 20

3.1 图像特征检测与描述子获取 ......................................................................... 20

3.1.1 SIFT 算法 ............................................................................................. 20

3.1.2 SURF 算法 ........................................................................................... 20

3.1.3 ORB 算法 ............................................................................................. 21

3.1.4 三种方案实验比较 .............................................................................. 23

3.2 图像匹配算法 ................................................................................................. 25