The main goal of this thesis will be to design and develop an automatic pipeline for conversion of large input datasets to game-ready 3D digital content in Unreal Engine. The input datasets will consist of point cloud data, obtained by laser scanning, and image data. RealityCapture software will be used to perform multi-view reconstruction and mesh triangulation from image and point cloud data. This software will be accessed using API to achieve full automation in data processing. Moreover, advanced segmentation algorithms will be implemented in the thesis to obtain segmentation of individual objects for their usage as game objects in Unreal Engine. The exported game objects will be also integrated into Carla platform, based on Unreal Engine, for car simulation rendering. The quality of the reconstruction will be evaluated on multiple datasets using user-oriented subjective evaluation and it will be compared with different reconstruction pipelines (e.g. using MeshLab or NeRFs). Optionally, new datasets may be captured using our Lab equipment at the university. The resulting reconstruction pipeline, together with segmentation algorithms, will serve as a basis for rapid creation of simulation-ready and game-ready 3D digital content. The thesis will be done in collaboration with company partner Magna. A payment of 2500€ is offered by the partner company for successful accomplishment of this master thesis.
Type: Master thesis (can be also combined with practicum to extend the scope of the research)
Supervisors: Peter Kán and Hannes Kaufmann