Abstract
Surface registration brings multiple scans into a common coordinate system by aligning their overlapping components. This can be achieved by finding a few pairs of matched points on different scans using local shape descriptors and employing the matches to compute transformations to produce the alignment. By defining a unique local reference frame (LRF) and attaching an LRF to shape descriptors, the transformation can be computed using only one match based on aligning the LRFs. This paper proposes a local voxelizer descriptor, and the key ideas are to define a unique LRF using the support around a basis point, to perform voxelization for the local shape within a cubical volume aligned with the LRF, and to concatenate local features extracted from each voxel to construct the descriptor. An automatic rigid registration approach is given based on the local voxelizer and an expanding strategy that merges descriptor representations of aligned scans. Experiments show that our registration approach allows the acquisition of 3D models of various objects, and that the local voxelizer is robust to mesh noise and varying mesh resolution, in comparison to two state-of-the-art shape descriptors.
Highlights
With advances in depth sensing technology, it is easy and flexible to acquire depth scans could of real objects, e.g., using the Kinect [1, 2]
Registration methods can be broadly classified as local registration approaches using iterative closest point algorithm (ICP), global registration approaches that search for the best aligning transform, and registration based on local shape descriptors
Tombari et al [10] proposed an signature of histograms of orientations (SHOT) descriptor by constructing a unique local reference frame (LRF) for a feature point and concatenating local histograms defined on each bin within a 3D spherical volume aligned with the LRF
Summary
With advances in depth sensing technology, it is easy and flexible to acquire depth scans could of real objects, e.g., using the Kinect [1, 2]. By defining a unique local reference frame (LRF) for each point using its support (i.e., the local shape around the point) and attaching the LRF to descriptors, just one pair of matched points can determine the transformation by aligning the three axes of their LRFs [10, 11]. This drastically reduces the search space for corresponding points (i.e., from at least 3 pairs to 1 pair), and increases the chance to find correct aligning transforms for the scans. An automatic surface registration approach is proposed to acquire 3D models from a set of input scans. Experiments show that our registration method can acquire 3D models of objects with varying shape complexity
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
