Depth scans acquired from different views may contain nuisances such asnoise, occlusion, and varying point density. We propose a novel Signature ofGeometric Centroids descriptor, supporting direct shape matching on the scans,without requiring any preprocessing such as scan denoising or converting into amesh. First, we construct the descriptor by voxelizing the local shape within auniquely defined local reference frame and concatenating geometric centroid andpoint density features extracted from each voxel. Second, we compare twodescriptors by employing only corresponding voxels that are both non-empty,thus supporting matching incomplete local shape such as those close to scanboundary. Third, we propose a descriptor saliency measure and compute it from adescriptor-graph to improve shape matching performance. We demonstrate thedescriptor's robustness and effectiveness for shape matching by comparing itwith three state-of-the-art descriptors, and applying it to object/scenereconstruction and 3D object recognition.
展开▼
