Seversky, Lee M.Yin, LijunH. P. A. Lensch and S. Seipel2015-07-092015-07-092010https://doi.org/10.2312/egsh.20101056This paper presents a novel and automatic approach for aligning range-scan data of objects exhibiting non-rigid, articulated motion using a cage-driven reduced deformable model. Reduced deformable models have previously been used for non-rigid registration. However, these approaches usually assume a model apriori or determine one in step with the registration which adds complexity. We choose a cage-based space deformation mapping as the reduced deformable model and formulate the scan alignment problem as a space deformation problem. This cagebased deformation mapping provides a compact deformation model which is inherently geometric. We seek the deformation of a source cage (and embedded geometry) that results in the best alignment of the source and target scans. The main advantage of our approach is that the reduced deformable model is constructed automatically from the underlying object geometry and is independent of the alignment procedure as it does not require explicit partitioning of the object into parts or the establishment of joints. Our alignment algorithm is completely automatic and does not require initial correspondences between the surfaces to be aligned.This paper presents a novel and automatic approach for aligning range-scan data of objects exhibiting non-rigid, articulated motion using a cage-driven reduced deformable model. Reduced deformable models have previously been used for non-rigid registration. However, these approaches usually assume a model apriori or determine one in step with the registration which adds complexity. We choose a cage-based space deformation mapping as the reduced deformable model and formulate the scan alignment problem as a space deformation problem. This cagebased deformation mapping provides a compact deformation model which is inherently geometric. We seek the deformation of a source cage (and embedded geometry) that results in the best alignment of the source and target scans. The main advantage of our approach is that the reduced deformable model is constructed automatically from the underlying object geometry and is independent of the alignment procedure as it does not require explicit partitioning of the object into parts or the establishment of joints. Our alignment algorithm is completely automatic and does not require initial correspondences between the surfaces to be aligned.10.2312/egsh.2010105693-96