Description
A Master of Science thesis in Mechatronics Engineering by Abdallah Adel Abdeen entitled, “Semantic based navigation and lane keeping”, submitted in April 2023. Thesis advisor is Dr. Shayok Mukhopadhyay. Soft copy is available (Thesis, Completion Certificate, Approval Signatures, and AUS Archives Consent Form).
Abstract
This thesis develops a novel method of robotic navigation and lane keeping system for the outdoor environment. The system uses semantic information, that is knowledge of a given map, and camera-based object detection of landmarks on the map. This allows any robot/vehicle to identify its approximate location on the map without using any beacon-based sensors, but only using semantic data obtained from a single RGB camera. This approach also does not use any estimation algorithm. The approach combines artificial intelligence-based object detection along with path planning algorithms, to provide a user a path from one point of the map to another. The lane keeping portion of the algorithm follows road lane markings until the directions from the semantic navigation algorithm leading the user/robot/vehicle to its destination. This system can be used for navigation onboard vehicles where the driving is done by a human, or the navigation system can be plugged into the lane keeping system of an autonomous vehicle, for achieving autonomous driving capabilities onboard a robot or an autonomous vehicle. This work presents results showing that future navigation tasks can be made less dependent on requiring a multitude of sensing and computing hardware, in environments where reliable and high-quality maps are already available. This has the potential to make navigation for autonomous driving in urban areas less expensive, as requiring a suite of LIDAR/RADAR, imaging, precision GPS sensors; and fusing all the data together – as prevalent on current autonomous vehicles, is very expensive. Additionally, robots that are reliant on GPS sensors are very reliant on their connection with satellites, which can often fail. Our proposed method aims to create a self-sustained system with reduced costs by relying on visual data obtained from an inexpensive camera and using image processing and artificial intelligence to achieve a visual positioning system. Object detection was used as the main backbone of this system and from within our tests, 8 false positives were detected out of 9000 images, which is a promising result for building detection.