Space Carving Algorithm with Enhanced Visual Quality & Memory Efficiency
이정, 김창헌 · 대한기계학회 춘추학술대회 · 2006
We propose an image-based modeling algorithm that improves the space carving algorithm in terms of visual quality and memory management. Previous voxel-carving approaches robustly reconstruct target objects, but they have suffered from voxel quantization and inefficient memory allocation of initial volume. By converting reconstructed voxels into polygonal cubes with triangular faces and applying texture mapping technique to those cubes, the quantization artifact is reduced so much. Instead of the previous static initial volume for space carving, we construct an initial volume adaptively and it avoids the memory allocation of many voxels in definitely empty region. Artifacts of polygonal aliasing and texture seam are dramatically reduced by our new geometric smoothing operation. Experiments show the texture-mapped results and surfel-rendered results those are sampled from the reconstructed polygonal objects.