GNSS NLOS Signal Detection In Urban Environments Using Fish-eye Images Classification And Segmentation Analysis

The goal of this thesis is to present a heuristic that mitigates the positioning degradation in urban canyons by identifying NLOS signals using fish-eye images and remove them from the final positioning calculation.Using a wide angle camera pointed towards the sky we can record the environment and detect the open sky region and different categories of obstacles.Than the satellites are projected onto the segmented image and detect NLOS signals that are obstructed by the obstacles.The thesis describes a holistic development pipeline used for creating the necessary sub-systems that helped achieve the stated goal from Data Collection sub-system,to Obstacle Detection sub-system and NLOS Projection and Detection sub-system.It also describes all the necessary devices needed for this task and how to set up all of them to ensure accurate data collection.The entire collection system consists in a fish-eye camera,GNSS receiver,two antennas,an IMU device and a central processing unit.All of these devices are mounted onto a van used for various research projects.The data collected from all these devices needs to be processed to achieve the desired results.Thus,the thesis shows the analysis of different methods to achieve a reliable synchronization result between the collected GNSS signal and the visual signal,from industry based solutions to a in house developed synchronization algorithm,and argues for the chosen solution.A couple of image segmentation techniques are compared to decide which one offers the most promising results.Previously,classical machine learning algorithms where proposed for this stage to segment the image into sky and obstacle class,but this thesis looks at state of the art deep learning solutions as they could handle more complex scenarios.An image processing method based on morphological operations used to enhance the segmentation results is also described.In the end the thesis shows the necessary steps to project the satellites onto the image plane and detect the NLOS signals.Proposed innovations and future research directions are discussed for each stage of the development pipeline.The results obtained from each subsystem is presented and their validity and correctness is analysed leading to the conclusions of this thesis.