华 中 科 技 大 学 硕 士 学 位 论 文
II
Abstract
Recently, with the development of robotics, its core technology - Simultaneous
Location and Mapping (SLAM) has also been widely concerned. In the field of visual
SLAM, with its advantages of simultaneous access to color images and depth images, RGB-
D camera is widely used in the solution for indoor SLAM. However, there are still many
problems to be solved urgently in the real-time and accuracy aspect of the current RGB-D
SLAM algorithm. Therefore, based on the original solution of RGB-D SLAM, this thesis
has made a much deeper study on the important parts of SLAM system.
Firstly, in the part of SLAM frontend, the different feature extraction algorithms and
different feature matching algorithms are compared and analyzed, and an improved
RANSAC algorithm based on the adaptive method is proposed and applied to the pose
estimation. By doing comparative experiments, it’s proved that proposed RANSAC
algorithm can effectively improve the accuracy, stability and efficiency of the pose
estimation.
Secondly, in the parts of SLAM backend and mapping, a key frame extraction method
combining the method of inliers determination and relative motion-size determination is
proposed. Based on the research of key frame extraction and loop-closure search, a global
pose optimization graph is established and solved by g2o (General Graph Optimization). At
the same time, this thesis compares and analyzes the different two forms of the map.
Then, based on the above study, the experimental platform of RGB-D SLAM system
is established and its experimental study in real environment is carried out to verify the
feasibility of the system proposal. What’s more, by doing the comparative experiments of
different algorithms in different parts based on benchmark dataset, the influence of different
algorithms to the result of RGB-D SLAM system is verified, which is of great significance
to promote the application of RGB-D SLAM research in daily life.
Finally, the thesis summarizes the whole paper and makes a prospect for the future
research direction.
Keywords
: RGB-D camera; Simultaneous Location and Mapping (SLAM); Random
Sample Consensus (RANSAC); pose estimation; key frame extraction
评论0