ORCID

0000-0002-9412-8464

Keywords

Simultaneous Localization and Mapping, Robotics, Systematic Evaluation, Intrinsic Camera Properties, Synthetic Data

Abstract

Simultaneous localization and mapping (SLAM) is the process of estimating the pose of a robot within an unknown environment. A variant, visual SLAM (V-SLAM), estimates the pose only using visual information from a camera. In recent years, the importance of V-SLAM has increased and expanded to applications such as robotics, virtual reality, augmented reality, and autonomous driving. Accurate estimation of poses is critical as these applications need to quickly perceive an environment and update its pose under complex scenarios. V-SLAM is composed of algorithms that utilize robot vision to distinguish landmarks in the environment in order to estimate the geometric position of the robot and its direction. This process is developed from algorithms that can extract features, match features with prior images, locate features from the past, correct the estimations, and map an unknown environment. A weakness of V-SLAM utilizing visual-only information is that any visual disruptions or errors will impact the estimates of the pose. These impacts are reflected as incorrect estimates, pose drifts, loop closure failures, or in extreme cases, complete failures of the SLAM. The extent of these errors is influenced by factors such as intrinsic camera properties, such as the resolution, frames per second, and field of view. V-SLAM errors are also influenced by lighting, environment, trajectory of the camera, or vibrations applied directly to the camera. In this dissertation, I systematically explore the effects of camera, environment, and motion properties on V-SLAM. My investigation identifies ideal intrinsic camera properties for resolution and field of view for V-SLAM. Next, I evaluate different environments under varying lighting to determine the best V-SLAM lighting and environments. I conclude by investigating different trajectories and vibrations on the camera and the performance of the V-SLAM.

Completion Date

2025

Semester

Fall

Committee Chair

Wu, Annie, McMahan, Ryan

Degree

Doctor of Philosophy (Ph.D.)

College

College of Engineering and Computer Science

Department

Computer Science

Format

PDF

Identifier

DP0029830

Document Type

Thesis

Campus Location

Orlando (Main) Campus

Subjects

Computer vision--Research; Robots--Motion--Research; Cameras--Calibration; Virtual reality--Research; Image processing--Research

Download

Share

COinS