Rogério Yugo Takimoto

High-speed point cloud matching algorithm for medical volume images using 3D Voronoi diagram

Several respiratory diseases, such as COPD and asthma, requires periodical checkups and past data comparison. While this kind of analysis is usually done by a medical expert, it depends greatly on the medical expertise and the image quality. Image registration, a technique which compares images volumes automatically using predefined computational algorithms, is a great tool to assist on diagnosis and disease surveillance. Most studies analyze the registration on 3D CT images slice-by-slice. However, by segmenting a 3D point clouds from the 3D CT volumes, it is possible to analyze the data in different and more accurate ways. This paper proposes a high speed algorithm improvement that calculates the rigid registration between two point clouds, adapting the Iterative Closest Point (ICP) algorithm to use 3D Voronoi diagrams for point correspondence determination, reducing the processing time greatly. A benchmark performance test is done with a point-by-point variation of the algorithm, showing that the proposed algorithm yield the same results with a considerable processing time reduction.

Automatic Epipolar Geometry Recovery Using Two Images

Abstract The analysis and recovery of the epipolar geometry is a crucial step to perform a 3D reconstruction of a scene. This work uses two uncalibrated images as input to compute the epipolar geometry of a scene. This is done in two steps: 1. automatically feature points extraction and 2. feature points mapping determination. The feature points from the two images are automatically extracted through the SIFT algorithm. The ICP algorithm is used to compute an initial correspondence among the feature points by comparing their associated information. A novel robust mapping determination algorithm is proposed to speed up the matching process while the accuracy is maintained. The main idea is that the order in which three visible feature points in the 3D are seen must be the same independently of the camera position. The Delaunay triangulation creates coherently oriented triangles from the obtained inliers. Inliers that defined non coherently triangles are removed. The convex hull of the Delaunay triangulation is used to determined a new set of 8 points and a new set of inliers is determined. The proposed algorithm was tested and showed to be robust. Copyright

3D Reconstruction Using Low Precision Scanner

The objective of this work is to use low precision laser sensor and create reasonably precise 3D reconstructions. The 3D reconstruction is executed by combining several point clouds obtained from different viewpoints. The proposed method was developed with three main steps: point cloud registration, error compensation and surface reconstruction. The ICP algorithm is improved to execute the point cloud registration: dynamic distance threshold, weighted distance, rigid body restriction and color information. It is shown that using this improved ICP, the number of point correspondences to evaluate the quadratic error converges to a value. The quadratic error can be determined independently of scene complexity. The point cloud errors are compensated using the consensus surface algorithm with signed distance. The surface is reconstructed using the marching cubes algorithm. Several results are shown to demonstrate the reliability of the proposed method.

Epipolar Geometry Estimation, Metric Reconstruction and Error Analysis from Two Images

Abstract The analysis and estimation of the epipolar geometry is a crucial step to perform a 3D reconstruction of a scene. In this work, the epipolar geometry of a scene was automatically recovered using two uncalibrated images. The feature points are determined by the SIFT and the feature mapping is executed by the ICP. The initial mapping has mismatches and is refined by the proposed method that considers the coherence between mapped triangles. The main idea is that the order in which three visible feature points in the 3D are seen must be the same independently of the camera position. This technique guides the process of choosing the set of points used to determine the epipolar geometry and to reject the inconsistent matches. The error analysis and metric reconstruction are also executed. Some results are shown and the proposed method is compared with the ground truth. Copyright

3D RECONSTRUCTION OF LARGE POINT CLOUDS WITH A NEW POINT CORRESPONDENCE ALGORITHM

The objective of this work is to perform the 3D reconstruction combining cloud points obtained from different viewpoints using structured light. The point cloud is simplified to reduce the computational time. The main task is the point cloud registration algorithm that matches two point clouds. A well known algorithm for point cloud registration is the ICP that determines the rotation and translation that when applied to one of the point clouds, place both point clouds in accordance. The ICP algorithm executes iteratively two main steps: point correspondence determination and registration algorithm. The point correspondence determination is a module that if not executed properly can make the ICP to converge to a local minimum. To overcome such drawback, it is proposed in this work an ICP that uses statistics to generate a dynamic distance threshold on the distance allowed between closest points. Instead of matching all points from the data set, this technique matches subset-subset points.

Propagation-based marching cubes algorithm using open boundary loop

The marching cubes (MC) algorithm is employed to generated triangular meshes for visualizing medical images, sculpture scans and mathematical surfaces. It sequentially traverses cuberille data composed of sampled points of a scalar volumetric data. This paper proposes a propagation-based MC algorithm that uses the open boundary loop concept. The open boundary loop is used to determine adjacent cells for the next iteration of the MC algorithm. After inserting each triangle, the open boundary loop is reevaluated. Simultaneously, it is ensured that all triangles are coherently oriented and there are no holes on the isosurface. Several tests are conducted to determine the performance of the algorithm in comparison with the original MC algorithm. Results from these tests indicate that, for large-scale problems, the proposed algorithm performs better than the original.

Shape reconstruction from multiple RGB-D point cloud registration

The objective of this work is to present an object 3D reconstruction method using the point color information. The object 3D reconstruction is performed by combining point clouds obtained from different viewpoints using two cameras and a structured light projector. The main task is the point cloud registration algorithm that matches two point clouds. A well known algorithm for point cloud registration is the ICP (Iterative Closest Point) that determines the rotation and translation that when applied to one of the point clouds, place both point clouds in accordance. The ICP algorithm executes iteratively two main steps: point correspondence determination and registration algorithm. The point correspondence determination is a module that if not executed properly can make the ICP to converge to a local minimum. To overcome such drawback an ICP that uses statistics to generate a dynamic distance and color threshold on the distance allowed between closest points is proposed and implemented. This approach allows subset matches, instead of matching all points from the point clouds. The surface reconstruction is performed using Marching Cubes and a consensus surface algorithm with signed distance compensates point cloud errors. In this paper the performance of the proposed method is analyzed and compared with the classical ICP.

Algorithmic iterative sampling in coordinate metrology plan for coordinate metrology using dynamic uncertainty analysis

Coordinate metrology is inherently subject to a source of uncertainty due to an attempt to inspect an unknown surface based on a limited number of discrete observations called sampling points. The computation tasks required for this evaluation need to be designed and conducted to minimize the uncertainty factors during the inspection process. This work presents a novel sampling planning approach based on a probabilistic framework to estimate the uncertainty in reconstruction of the measured surface. The goal is to minimize the required number of sample points to inspect a surface flatness within an acceptable level of uncertainty. The developed methodology models the deviation from the ideal geometry is modeled as a linear combination of shape functions. Then a Probability Density Function (PDF) is created based on a prior model of the expected surfaces deviation characteristics. By combining the prior probability density function and the current set of measurements, a new PDF for the reconstructed deviation is updated during the measurement process which which combines their expected values and their uncertainties. This PDF in turn can be used to estimate critical points for flatness measurement. Those critical points are in turn elected to be sampled at the next measurements. The proposed adaptive sampling is evaluated using virtual sampling of a machined surface. Results show important improvement over the commonly used random sampling approaches.

Translational placement using simulated annealing and collision free region with parallel processing

The irregular shape packing problem is a combinatorial optimization problem that consists of arranging items on a container in such way that no item overlaps. In this paper we adopt a solution that places the items sequentially, touching the already placed items or the container. To place a new item without overlaps, the collision free region for the new item is robustly computed using non manifold Boolean operations. A simulated annealing algorithm controls the items sequence of placement, the items placement and orientation. In this work, the placement occurs at collision free regions vertices. Several results with benchmark datasets obtained from the literature are reported. Some of them are the best already reported in the literature. To improve the computational cost performance of the algorithm, a parallelization method to determine the collision free region is proposed. We demonstrated two possible algorithms to compute the collision free region, and only one of them can be parallelized. The results showed that the parallelized version is better than the sequential approach only for datasets with very large number of items. The computational cost of the non manifold Boolean operation algorithm is strongly dependent on the number of vertices of the original polygons.

Mesh Generation for Surfaces with Distinct Boundary Segmentation

The mesh approximation of solid models is an important and necessary step for several applications. Particularly when surfaces with T-junctions are presented in the model. In this situation, the surface boundaries can have different number of segment approximations. Such a surface can be filled by an irregular mesh. The paving algorithm is one of the best mesh generators, however its implementation is not easy. This work redesign the paving algorithm with a specific implementation to mesh quadrilaterals with different segmentaion in each side. Several results are shown demonstrating that the redesigned paving algorithm was successfully implemented.