哈尔滨工业大学工学硕士学位论文
II
Abstract
With the development of artificial intelligence and robotics technology,
robots play a more and more important rule in people's lives. For mobile robots,
localization has always been a core capability. The purpose of This article is to
use the IMU (Inertial Measurement Unit) to improve the performance of the
visual SLAM (simultaneous localization and mapping).
First of all, this article introduces the camera's pinhole model, binocular
vision model, binocular distance measurement principle and other basic
knowledge of binocular vision. Then FAST feature points, BRIFF descriptors,
and ORB feature points is introduced. Afterwards, the method of loop detection
is introduced.
Secondly, this article establishes a noise model and a kinematics model for
the IMU, and pre-integrates the measured quantities of the IMU. And the noise
propagation equation is derived and the influence of random walk is analyzed.
Then, in this paper, IMU information and visual information are combined
using nonlinear optimization methods A method of saving the map and how to
use the map for more precise and fast localizatio is also given later .
Finally, experiments are designed and conducted based on the proposed
algorithm. Among them, the calibration experiment calibrated the parameters of
camera and the external parameters of the camera and the IMU. Experiments
using the Euroc dataset compare the error between the localization results of this
algorithm and the ground truth. Finally, the map of some real scene around us is
build using the mynt-eye binocular camera.
Keywords: localization, visual SLAM, IMU, information fusion, map saving
and reuse
评论0