3D free-form surface registration and object recognition

Abstract

A new technique to recognise 3D free-form objects via registration is proposed. This technique attempts to register a free-form surface, represented by a set of 21D sensed data points, to the model surface, represented by another set of 21D model data points, without prior knowledge of correspondence or view points between the two point sets. With an initial assumption that the sensed surface be part of a more complete model surface, the algorithm begins by selecting three dispersed, reliable points on the sensed surface. To find the three corresponding model points, the method uses the principal curvatures and the Darboux frames to restrict the search over the model space. Invariably, many possible model 3-tuples will be found. For each hypothesized model 3-tuple, the transformation to match the sensed 3-tuple to the model 3-tuple can be determined. A heuristic search is proposed to single out the optimal transformation in low order time. For realistic object recognition or registration, where the two range images are often extracted from different view points of the model, the earlier assumption that the sensed surface be part of a more complete model surface cannot be relied on. With this, the sensed 3-tuple must be chosen such that the three sensed points lie on the common region visible to both the sensed and model views. We propose an algorithm to select a minimal non-redundant set of 3-tuples such that at least one of the 3-tuples will lie on the overlap. Applying the previous algorithm to each 3-tuple within this set, the optimal transformation can be determined. Experiments using data obtained from a range finder have indicated fast registration for relatively complex test cases. If the optimal registrations between the sensed data (candidate) and each of a set of model data are found, then, for 3D object recognition purposes, the minimal best fit error can be used as the decision rule.

Three-dimensional object registration using wavelet features

“…Chua and Jarvis [10] present an algorithm which determines full and partial matches of surfaces represented by point cloud data. They determine principal curvatures and directions to form a Darboux frame.…”

“…Hausdorff Distance [2] 2003 NR Darboux Frames [10] 1996 70-153 sec Feature Point Mesh [93] 2003 NR Footprints [3] 1997 73-340 sec Point Fingerprint [82] 2003 NR KH Method [38] 2002 225 sec Lines of 0 H [40] 2002 NR Splash [77] 1992 35-1800 sec Umbilics [39] 2003 NR Dynamical Systems [16] 2003 NR Hierarchical Patches [75] 2006 NR Spin Images [33] 1997 415 sec Surface Signatures [99] 2002 120 sec WAV [98] 2003 NR Combination [85] 2006 1600-2200 sec Voxel Intensity Methods Mutual Information [88,95] 1995 400 sec Hardware Accelerated [23] 1998 400 sec Cumulative Distribution [89] 2003 NR Gradient Descent [6] 2000 150 sec PET/CT Auto Elastic [70] 2005 2700-4500 sec Q-MI [44] 2008 NR Block Match [47] 2006 1000 sec Table 2.2: Speed of registration methods. The times listed are those reported in the literature; NR indicates that computation time was not reported.…”

Deformable object tracking using optical flow constraints

“…Image-based feature points are e.g. local curvature extrema or saddle points, edges or corners [5], [20], [11] or SIFT features [15]. Popular choices of shape-features are the shape context approach of Belongie et al [2], or statistical moments of the shape [4].…”

Matching Tensors for Automatic Correspondence and Registration

“…Moreover, bitangent curves are global features and may not be fully contained inside the overlapping region of the views. Three tuple matching [7] calculates the first and second order derivatives which are sensitive to noise and require the underlying surfaces to be smooth. SAI matching [8] requires the underlying surfaces to be free of topological holes.…”