| dc.description.abstract | For many decades researchers and manufacturers have tried to push the concept of augmented reality toward a comfortable, beneficial, and seamless experience. The concept of this technology promised high usability for fields such as medicine, architecture, and engineering, while keeping the experience smooth enough to avoid common negative effects such as nausea and dizziness. One of the most fundamental things towards realistic AR experience is occlusion by letting our virtual world and holograms be affected by the real world that surrounds us. In this Master's thesis I discover methods to apply occlusion in AR and the computational challenges that it brings. I focus on mesh-based reconstruction algorithms and consider generating meshes by getting insight into older algorithms like Marching Cubes and Poisson, as well as newly developed solutions. I answer a research question by creating our own mesh-based reconstruction pipeline: 'Can we develop efficient real-time, mesh-based method to occlude virtual objects with real objects in an AR scenario?'. I have concluded that developing such a solution is possible within a reasonable amount of time using Unity Engine and Marching Cubes, yielding good results in stereo-camera setup on modern hardware. The performance of the developed solution was measured using Unity Profiler and custom measurement tools. The findings indicate that the development of a mesh-based solution should be heavily based on GPU computational power and minimize data movement. In addition, AR scene management, the LoD system, memory pooling, and work time slicing are factors that affect performance and resource usage. Statistics show that handling and dispatching the world in larger sectors is significantly better for performance. The challenge of cached approach and marching cubes is its memory wastefulness and dealing with moving objects. This can be improved by developing solutions for smart section discarding, mesh optimization pass, or different mesh generation algorithms. From the research, we can conclude that AR technology has cultivated over the decades and has blossomed in recent years, but many areas, such as occlusion, require more research. | eng |