Recently, quite a number of approaches came up to reconstruct 3D volumes from RGB or RGBD camera input. However, most of these approaches are rather inflexible regarding the initial camera position with respect of the reconstruction volume and the overall size of the area to be reconstructed. This severly limits the usability of those approaches. In this work we present a flexible approach to store and dynamically extend the reconstruction volume overcoming those problems. We show that our approach additionally requires significantly less memory due to a pyramid-based data storage. We demonstrate that our approach is real-time capable when implemented using the GPU and by that provides a flexible alternative to data structures used in previous approaches.
展开▼