Developing structural constraints for accurate registration of overlapping range images

Automatic registration of overlapping 3D point clouds using closest points

“…At the beginning of registration where the camera motion parameters are not accurate and occlusion and appearance and disappearance of points are present, the CPC establishes only pseudo-correspondences, since in a strict sense, none of them may represent the same physical points in 3D space. In this case, it is difficult to set up a threshold, indicating which are real correspondences and which are false correspondences, as is the case for the algorithms proposed in [22,27,28,29,30,31,57], including the GICP algorithm, due to the inherently continuous change of relative parameters used for classification. The mean field theory inspires us that a compromise can be made.…”

“…While the row normalisation is relatively easy to implement, the column normalisation needs special attention: even though a row has just two columns, we have to process them into n 2 columns, corresponding to each point in the second view. Comparing either the SoftICP or WeightICP algorithm with the existing improved ICP algorithms described in the introduction, the former has a clear advantage of easy implementation and is of general use, since they do not require any feature extraction [20,44,54], parameter estimation [31,27,41] necessary for the elimination of false correspondences, or any structural information from data points [14,51].…”

“…This is because the former did not completely eliminate false matches, accelerating the registration process [27], which often converges to a local minimum, yielding poor registration results as already demonstrated above.…”

“…As a result, the key to successfully applying the ICP algorithm for 3D free form shape registration lies in eliminating the correspondences with relatively lower qualities. For this purpose, the following techniques have been proposed: † Increase the dimensionality of points from 3D [3,4] to higher dimensions by incorporating other geometric or optical features, such as normal vectors [12], invariants [44], curvature [55], laser reflectance strength value [35], and colours [22]; † Sample points uniformly [51], in normal space [42], or based on covariance matrix [14]; † Establish correspondences from matching points to matching curves [48,53] to matching 2D images [20,23,54], from matching local structural features to examining motion consistency [28,31,41] to combining both [27]; † Consider the reliability of point correspondences as a function of the cosine of the including angle between vertex normals and their viewing directions [51]; † Eliminate false matches through removing boundary points [42,51], checking interpoint distance consistency [11] or orientation consistency [38,57], examining motion consistency [29][30][31]41], or examining both the motion and structural consistency [28,27]; † Estimate motion parameters from using least squares to using weighted least squares [51], genetic algorithm [45], M-estimator [35] or simulated annealing [33], or from the Euclidean space to the frequency space [32].…”

Replicator Dynamics in the Iterative Process for Accurate Range Image Matching

“…As a result, a large number of ICP variants have been proposed for more accurate and robust matching results based on techniques such as removing boundary points (Turk and Levoy 1994); using rigid motion constraints (Liu et al 2000) or collinearity constraints (Liu et al 2006a); combining rigid motion constraints with structural constraints (Liu and Wei 2004); checking the interpoint distance compatibility (Dorai and Wang 1998) or orientation consistency (Pulli 1999;Zhang 1992); increasing the dimensionality of points using invariant features (Sharp et al 2002), colour (Schutz et al 1998), or laser reflectance strength value (Allen et al 2003); matching interpoint distances (Chen et al 1999), 2D spin images Huber and Hebert 2003;Johnson and Hebert 1999), textured 2D spin images (Brusco et al 2005), or surface signatures (Yamany and Farag 2002); resampling points in the normal space (Rusinkiewicz and Levoy 2001); dynamically thresholding false correspondences (Zhao et al 2004); using reciprocal points for the camera motion estimation (Pajdla and Van Gool 1995); bootstrapping camera motion models and transformation estimate (Stewart et al 2003); and taking the confidence value of each point into account for camera motion estimation (Zhu et al 2007). In order to update the camera motion parameters, it is vital to evaluate the possible correspondences established.…”