Computational Analysis of Distance Operators for the Iterative Closest Point Algorithm

Abstract

The Iterative Closest Point (ICP) algorithm is currently one of the most popular methods for rigid registration so that it has become the standard in the Robotics and Computer Vision communities. Many applications take advantage of it to align 2D/3D surfaces due to its popularity and simplicity. Nevertheless, some of its phases present a high computational cost thus rendering impossible some of its applications. In this work, it is proposed an efficient approach for the matching phase of the Iterative Closest Point algorithm. This stage is the main bottleneck of that method so that any efficiency improvement has a great positive impact on the performance of the algorithm. The proposal consists in using low computational cost point-to-point distance metrics instead of classic Euclidean one. The candidates analysed are the Chebyshev and Manhattan distance metrics due to their simpler formulation. The experiments carried out have validated the performance, robustness and quality of the proposal. Different experimental cases and configurations have been set up including a heterogeneous set of 3D figures, several scenarios with partial data and random noise. The results prove that an average speed up of 14% can be obtained while preserving the convergence properties of the algorithm and the quality of the final results.

A novel rotation invariant and Manhattan metric–based pose refinement: Augmented reality–based oral and maxillofacial surgery

“…This feature also decreases the number of iterations needed for ICP; therefore, it accelerates the performance of the image registration process. In addition, we added Manhattan error metric, which was proposed by Mora et al to reduce the processing time. It helps to improve the processing time and has the same accuracy as a Euclidian distance metric.…”

“…Furthermore, Table shows the rotation invariant based (global reference point) ICP algorithm's steps. Moreover, the Manhattan metric is used for calculation of the error between two point sets, which requires lower computational cost than the Euclidian distance error metric. The Euclidian distance error metric is only possible with two straight points and lines, whereas the Manhattan error metric is capable of calculating any type of pixels and at any angle.…”

“…The Euclidian distance error metric is only possible with two straight points and lines, whereas the Manhattan error metric is capable of calculating any type of pixels and at any angle. The use of the Manhattan metric instead of the Euclidean one will improve the processing time of the system . Then the final refined 3D pose is obtained and overlaid onto the real‐time video to create AR view in the operation room.…”

“…To improve the processing time of the distance calculation between the two point sets, the Manhattan error metric is employed instead of Euclidian error metric, presented as Equation where…”

An Adaptive △M-ICCP Geomagnetic Matching Algorithm

“…In fact, many factors affect the matching length and it is difficult to theoretically analyse the influences of these factors. To deal with the contradictions between matching precision and real-time performance, a point-by-point iteration matching algorithm (Mora et al, 2016) and iterative evaluation matching algorithm (Huang et al, 2012) are proposed. Their main idea is to judge whether the matching results meet the given criteria when the matching length increases.…”