An Improved Robot’s Localization and Mapping Method Based on ORB-SLAM
Hailan Kuang, Xiaodan Wang, Xinhua Liu *, Xiaolin Ma and Ruifang Li
School of Information Engineering
Wuhan University of Technology
WuHan, China
kuanghailan@whut.edu.cn (H.K.); wxxd@whut.edu.cn (X.W.); maxiaolin0615@whut.edu.cn (X.M.); liruifang@whut.edu.cn
(R.L.)
Abstract— The ORB-SLAM is a real-time SLAM system based
on feature points, with high accurate positioning, high robust,
and the ability to operate in large-scale, small-scale, indoor and
outdoor environments. However, since the whole SLAM
system uses feature points and each image needs to calculate
ORB characteristics once, which will consume a lot of time.
Therefore, using quasi-physical sampling algorithm based on
BING features combined with depth information to preprocess
image. And the matching strategy of the ORB algorithm is
optimized by an improved KD-Tree to improve the matching
speed on the premise of invariable matching accuracy. Finally,
on the basis of the above, using the KINECT to construct the
3D dense point cloud map system, improving the deficiency of
ORB-SLAM which can only build sparse map. The results of
experiment verified the feasibility of the algorithm, and can be
robust and fast in complex indoor environments, which can be
applied to the positioning and build 3D dense map.
Keywords-SLAM; ORB; KINECT; 3D dense map
I. INTRODUCTION
When the robot moves in an unknown environment, how
to determine its own trajectory through observation of the
environment, and constructing a map of the environment,
which is the Simultaneous Localization and Mapping
problem [1]. It has become the research hotspot in the field
of mobile robot. According to the solving process of SLAM,
it can be divided into two categories: the method based on
the filter and the method based on the graph optimization.
SLAM method based on the filter uses the recursive
Bayesian estimation, to construct the map incrementally and
to realize localization, mainly including Extended Kalman
Filter (EKF) and the Particle Filter (PF). EKF-SLAM was
first proposed by Smith et al. [2], the position and feature
map of the mobile robot is represented by a joint state vector,
and then EKF is used to estimate the mean and variance of
the state vector. However, EKF-SLAM is no longer
applicable when systems are non-linear. Therefore, Julier et
al. [3] proposes the Unscented Kalman Filter (UKF) method.
But both EKF and UKF have the problem of data correlation
sensitivity and high computational cost, so they are no longer
widely used. So Sim et al. [4] is introduced into the Particle
Filter (Particle Filter, PF), the idea is to use the distribution
of the particles to express the robot position and road signs
point, which avoids the complicated calculation of KF and
the error sensitivity of data association. But it still faces high
computational costs due to the large number of particles and
the depletion of particles due to heavy sampling.
With the development of the sparse matrix and nonlinear
optimization theory, SLAM method based on graph
optimization [5] arises at the historic moment, is essentially
based on the principle of smooth. Through all the
observation information, to estimated the robot's complete
motion trajectory and constructed the environmental map,
which becomes the mainstream method of visual SLAM
nowadays. In 2007, Klein and Murray in the papers [6]
proposed the Parallel Tracking and Mapping (PTAM)
method, which realized the parallelization of tracing and
construction process. It was the first online real-time SLAM
system to be realized by graph optimization. After that, Mur-
Artal et al. [7] proposed the ORB-SLAM algorithm, the
algorithm contains trace, map building, relocation of three
parallel threads, which can achieve high accuracy by online
real-time in a small scale and wide range of unknown
environments.
However, there are some drawbacks of the ORB-SLAM.
Due to the whole SLAM system uses feature points to
calculate, each image needs to calculate once ORB
characteristics, which is very time consuming. And we know
that the mapping of the ORB-SLAM is a sparse feature
point. The sparse feature point map can only meet the
requirements of our positioning, and can't provide many
functions such as navigation, obstacle avoidance and
interaction.
II. T
HE ALGORITHM INTRODUCED IN THIS PAPER
Depth
RGB
Ima ge
Regional
Block
ORB
Feature
Extraction
Improved
ORB
Matching
Algorithm
RANSAC
Robot
Posture
δPN Pε
Posture
Optimization
3D Point
Cloud
Model
3D Dense
Map
KIN ECT
Figure 1. Flow chart of the algorithm in this paper.
Aiming at the problems existing in the ORB-SLAM, the
optimization of this paper proposed is mainly the 3D dense
point cloud map system based on KINECT. And takes quasi-
physical sampling algorithm based on BING features
combined with depth information to preprocessed image,
2017 10th International Symposium on Computational Intelligence and Design
2473-3547/17 $31.00 © 2017 IEEE
DOI 10.1109/ISCID.2017.179
400
下载后可阅读完整内容，剩余3页未读，立即下载
weixin_38625442
- 粉丝: 6
- 资源: 950
我的内容管理
展开
最新资源
- 十种常见电感线圈电感量计算公式详解
- 军用车辆：CAN总线的集成与优势
- CAN总线在汽车智能换档系统中的作用与实现
- CAN总线数据超载问题及解决策略
- 汽车车身系统CAN总线设计与应用
- SAP企业需求深度剖析：财务会计与供应链的关键流程与改进策略
- CAN总线在发动机电控系统中的通信设计实践
- Spring与iBATIS整合：快速开发与比较分析
- CAN总线驱动的整车管理系统硬件设计详解
- CAN总线通讯智能节点设计与实现
- DSP实现电动汽车CAN总线通讯技术
- CAN协议网关设计：自动位速率检测与互连
- Xcode免证书调试iPad程序开发指南
- 分布式数据库查询优化算法探讨
- Win7安装VC++6.0完全指南：解决兼容性与Office冲突
- MFC实现学生信息管理系统：登录与数据库操作
资源上传下载、课程学习等过程中有任何疑问或建议，欢迎提出宝贵意见哦~我们会及时处理！
点击此处反馈
