3D Gaussian Splatting i GIS: Metod och tillämpningar : En kombinerad litteraturstudie och ett pilotprojekt baserat på Postshot och ArcGIS Drone2Map

Sammanfattning

3D reconstruction is a key component of modern GIS applications, but traditional mesh-based methods suffer from limitations in performance, accuracy, and realism. This study aims to evaluate 3D Gaussian Splatting (3DGS) as an alternative approach and compare it with mesh-based reconstruction and Neural Radiance Fields (NeRF). The research combines a literature review with a practical pilot study, where UAV-based imagery is processed and analysed using Postshot and ArcGIS Drone2Map. The results show that 3DGS produces higher visual quality and more detailed geometry than mesh-based models, although at the cost of increased computational requirements and larger input datasets. Furthermore, the findings indicate that 3DGS has strong potential within GIS applications such as visualization, digital twins, and real-time rendering, while limitations remain for direct analytical use. In conclusion, 3DGS emerges as a promising technique that can complement or partially replace existing methods, while potentially contributing to a technological breakthrough in 3D modelling and enabling new applications of 3D within GIS.