الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 105 |  |  |
| | | from | 105 | | |
Abstract
Simultaneous Localization and Mapping (SLAM) is a computational approach for creating a map while simultaneously localizing a moving item in an uncertain environment, such as a car or maybe even a robot. Researchers included several sorts of sensors, which included laser sensors, encoders, inertial measurement units (IMU), Global positioning, and a wide variety of cameras, within early SLAM algorithms. However, SLAM algorithms are still suffering when used under low light intensities. The maturity of current SLAM algorithms has been increasing in the past year; however, low light conditions make computer vision practices difficult, as it is difficult to have accurate maps with a moving object being localized in it with challenging low light conditions like the ones encountered when dealing with underwater images. In this research, we look at how pre-processing might increase the accuracy and performance of simultaneous localization and mapping in low-light situations, particularly those shot underwater. This thesis examines the effects of traditional and deep learning pre-processing methods. The conventional contrast limited adaptive histogram equalization strategy achieves the highest results after hyperparameter adjustment, with a 20.74% improvement in accuracy on the Aqualoc underwater dataset.