ABSTRACT
Simultaneous Localization and Mapping（SLAM）is a hot topic in Robotics and
Computer Vision communities regions, and for emerging technologies such as
autonomous cars, augmented or virtual reality, service robots and drone navigation
SLAM is in great need. Cameras provide rich information of the environment, while
IMU(Inertial Measurement Unit) sensors provide self-motion information. Visual SLAM
is sensitive to rapid rotation, while IMU measure data with drift. A monocular camera has
the problem of scale ambiguity, while IMU can provide data in metric scale. Meanwhile
both the monocular camera and while IMU are cheap and useful. So monocular camera
and IMU fusion maybe a perfect solution for future SLAM.
In recent years there have been excellent results in Visual-Inertial SLAM, which
can estimate accurately the state of robot and simultaneously build a map of the
environment. However these approaches lack the capability to close loops and reuse map,
and trajectory estimation accumulates drift even if the sensor is continually revisiting the
same place. In this work we propose a novel Visual-Inertial SLAM initialization method,
which can compute the scale, the gravity direction, the velocity, and gyroscope and
accelerometer biases with high accuracy. Then base on this initialization method, we
present a novel tightly-coupled Visual-Inertial Simultaneous Localization and Mapping
system that can be able to close loops and reuse its map. Test results verify that our
Visual-Inertial SLAM achieve a typical scale factor error of 2\% and centimeter precision.
In contrast to other Visual-Inertial SLAM approaches, our Visual-Inertial SLAM can
close loop and reuse its map, so that there is no error accumulation when revisiting same
scene.
Keywords: Visual SLAM，IMU，tightly-coupled，Initialization，close loop，map reused
评论0