GPU‐Based Volumetric Reconstruction and Rendering of Trees From Multiple Images

Abstract

This paper presents a new hardware‐accelerated approach to the volumetric reconstruction of trees from photographs, based on the methods introduced by Reche‐Martinez et al. The system applies an adapted computed tomography (CT) procedure that uses a set of oriented photographs with known interior and exterior camera parameters for creating a 3D model of a tree, while requiring considerably fewer images than standard CT. As tomographic reconstructions are complex tasks that result in time‐consuming processes for high‐resolution volumes, the hitherto existing methods are improved and modified to allow an ideal parallelisation of the computations on graphics hardware. The paper delivers a detailed insight into the complete process of the reconstruction, from the acquisition and preparation of the input data to the implementation of the final system on graphics processing hardware.

Tree modelling from mobile laser scanning data‐sets

“…Then the images are projected from image to object space in a voxel representation. For each voxel a billboard texture is created for view‐dependent rendering (Shlyakhter et al, 2001; Reche‐Martinez et al, 2004; Vock et al, 2010). Voxel approaches are also used for tree model reconstruction from ALS (Wang et al, 2008).…”