TY  - JOUR
T1  - Evaluation of a Proposed Blunder Detection Method in ICP Algorithm for
Fine Registration of Point Cloud Data
AU - Ranjbar, M. AU - Saadatseresht, M. 
JO  - Journal of Engineering and Applied Sciences
VL  - 12
IS  - 7
SP  - 1902
EP  - 1909
PY  - 2017
DA  - 2001/08/19
SN  - 1816-949x
DO  - jeasci.2017.1902.1909
UR  - https://makhillpublications.co/view-article.php?doi=jeasci.2017.1902.1909
KW  - Registration
KW  -thersholding
KW  -normal vector
KW  -point cloud
KW  -laser scanning
KW  -SVD
KW  -quaternion
AB  - Laser scanners directly measure 3D coordinates of huge amounts of points in a short time period, so that it has a well-known solution in 3D object modeling. The abundant data of laser scanner can be efficiently utilized to model the scene, however in many cases; the object has to be scanned from different viewpoints due to accuracy occlusion field of view and range limitations. Because each scan has its own local coordinate system, all the different point clouds must be transformed into a common coordinate system. This procedure is usually referred to ‘registration’. There are many methods for registration problem including Target based, Image based and surface based registration methods. Surface based registration techniques give the highest registration accuracy and automation. This dissertation addresses refinement issue of surface based registration by use of an accurate and fast ICP algorithm. In the ICP algorithm, every point in one surface should be matched to a point on the other surface so that the matched surfaces have minimum deflection error. In our research, we compared both distant and normal vector thersholding and proposed a combined method in 3D surface matching. Analysis and experimental results demonstrated the proposed combined method gives better registration accuracy than the other standard approaches. Also in the research, different factors in accuracy and efficiency of registration algorithm are tested such as mathematical criterion function, using respectively Singular Value Decomposition (SVD) or eigen system computation based on the standard (R, T) representation and the eigen system analysis of matrices derived from quaternion forms of the transform, simple/complex surface geometry, initial values, surface overlapping and data conditions (ideal or noisy). The examinations demonstrate that accuracy of registration is not significantly different for the both mathematical computation methods of SVD and quaternion. It also shows that efficiency of registeration diminish for tough breaking surfaces and by low approximation of initial values. Registration has higher accuracy for more overlapping surfaces and lower level of noise in data. Therefore determination of initial overlap area especially for noisy data is so important.
ER  -