Research‎ > ‎Navigation‎ > ‎

Visual SLAM

Visual SLAM using a RGB-D Sensor


- Research Background

Visual SLAM using RGB-D sensor has many advantages over ceiling SLAM using mono camera.

There are many useful features in indoor environment (corner, line, plane, etc.).

Various features can be used as landmarks for visual SLAM.

- Research Objectives

Various feature extraction using a RGB-D sensor as landmarks of visual SLAM.

Robust visual SLAM in various indoor environments.

Accumulated error correction using loop closure.

Implemented and optimized for embedded system.


- Research Output

Keyframe based Visual Odometry.

Loop closure using pose graph.

Visual SLAM in indoor environment

- DeResearch Background

* Forward-looking RGB-D sensor based SLAM

Visual Inertial SLAM using a Camera (KU-vSLAM Ver. 2)


- Research Background

KU-vSLAM ver.2 employs visual inertial odometry (VIO) which uses a stereo camera and IMU for an alternative ego-motion estimation besides wheel odometry.

Various features besides the points are used to VIO measurement improve the robot pose estimation. .

Improved ego-motion estimation can reduce drift error --> less loop-closure is needed.

Keyframes are selected within the VIO trajectories to construct factor graphs and optimized to perform the loop-closure.

- Research Objectives

Robust visual inertial SLAM even in large environments.

Various feature extraction using a camera as landmarks for visual inertial odometry (VIO) in KU-vSLAM ver.2.

Stable feature track using an IMU.

Implemented and optimized for embedded system.

- Research Output

Modularized VI SLAM developed in KU-vSLAM ver.2 can be easily used for various applications besides mobile robots since it does not depend on the wheel odometry (or any other hardware mounted odometers).