Dongjoe Shin

Clique descriptor of affine invariant regions for robust wide baseline image matching

Assuming that the image distortion between corresponding regions of a stereo pair of images with wide baseline can be approximated as an affine transformation if the regions are reasonably small, recent image matching algorithms have focused on affine invariant region (IR) detection and its description to increase the robustness in matching. However, the distinctiveness of an intensity-based region descriptor tends to deteriorate when an image includes homogeneous texture or repetitive pattern. To address this problem, we investigated the geometry of a local IR cluster (also called a clique) and propose a new clique-based image matching method. In the proposed method, the clique of an IR is estimated by Delaunay triangulation in a local affine frame and the Hausdorff distance is adopted for matching an inexact number of multiple descriptor vectors. We also introduce two adaptively weighted clique distances, where the neighbour distance in a clique is appropriately weighted according to characteristics of the local feature distribution. Experimental results show the clique-based matching method produces more tentative correspondences than variants of the SIFT-based method.

Progressively weighted affine adaptive correlation matching for quasi-dense 3D reconstruction

Correlation matching has been widely accepted as a rudimentary similarity measure to obtain dense 3D reconstruction from a stereo pair. In particular, given a large overlapping area between images with minimal scale differences, the correlation results followed by a geometrically constrained global optimisation delivers adequately dense and accurate reconstruction results. In order to achieve greater reliability, however, correlation matching should correctly account for the geometrical distortion introduced by the different viewing angles of the stereo or multi-view sensors. Conventional adaptive least squares correlation (ALSC) matching addresses this by modifying the shape of a matching window iteratively, assuming that the distortion can be approximated by an affine transform. Nevertheless, since an image captured from different viewing angle is often not practically identical due to scene occlusions, the matching confidence normally deteriorates. Subsequently, it affects the density of the reconstruction results from ALSC-based stereo region growing algorithms. To address this, we propose an advanced ALSC matching method that can progressively update matching weight for each pixel in an aggregating window using a relaxation labelling technique. The experimental results show that the proposed method can improve matching performance, which consequently enhances the quality of stereo reconstruction. Also, the results demonstrate its ability to refine a scale invariant conjugate point pair to an affine and scale invariant point pair.

Point to Point Calibration Method of Structured Light for Facial Data Reconstruction

Since the calibrating point pairs of a structured light are not easy to obtain, most previous work on calibration is related to the uncalibrating method. This paper proposed a new method for determining a set of 3D to 2D point pairs for the offline calibration of structured light system focused on 3D facial data acquisition for the recognition. The set of point pairs is simply determined based on epipolar geometry between a camera and structured light plane, and a structured light calibrating pattern. The structured light calibrating process is classified into two stages: the 3D point data acquisition stage and the corresponding 2D data acquisition stage. After point pairs are prepared, the Levenberg-Marquardt (LM) Algorithm is applied. Euclidian reconstruction can be achieved simply using a triangulation, and experimental results from simulation are presented.

ExoMars Rover PanCam: Autonomous & Computational Intelligence [Application Notes]

As a part of the Aurora programme for Mars exploration, funded by the United Kingdom Space Agency (UKSA) and European Space Agency (ESA), the UK contributes to the Exobiology on Mars (ExoMars) rover science and engineering programme, with a scheduled launch in 2018; Hence, our Panoramic Camera (PanCam) [9][15] research and development (R&D) is timely. PanCam consists of two stereo Wide Angle Cameras (WAC) and one High Resolution Camera (HRC). While the development is still ongoing, we used funding awarded by the University College London (UCL) Graduate School to conduct investigations in the Himalayas and at Mount Everest Base Camp (EBC), according to the ExoMars rover Reference Surface Mission (RSM). The investigations included capturing stereo and high resolution images using stereo WAC emulators and HRC emulator at altitudes 3490 m, 5150 m and above. Images from different WAC filters, and color images from HRC were acquired at various Pan and Tilt Unit (PTU) mast positions. Our investigation results show significant reduction in data volume with minimum loss in image quality. Furthermore, we introduce a novel autonomous and computational intelligent system called Mission-Specific Data Processor (MSDP) for the rover. It includes Pan-Cam, Visual Data Fusion (VDF), Learning-enabled Object Detection (LOD), Self-Learning Agent (SLA) [22], and Environment Model Library (EML) as part of the rovers computational intelligence [7].

Similarity Invariant Delaunay Graph Matching

Delaunay tessellation describes a set of arbitrarily distributed points as unique triangular graphs which preserves most local point configuration called a clique regardless of noise addition and partial occlusion. In this paper, this structure is utilised in a matching method and proposed a clique-based Hausdorff Distance (HD) to address point pattern matching problems. Since the proposed distance exploits similarity invariant features extracted from a clique, it is invariant to rotation, translation and scaling. Furthermore, it inherits noise robustness from HD and has partial matching ability because matching performs on local entities. Experimental results show that the proposed method performs better than the existing variants of the general HD.

Local Hull-Based Surface Construction of Volumetric Data From Silhouettes

The marching cubes (MC) is a general method which can construct a surface of an object from its volumetric data generated using a shape from silhouette method. Although MC is efficient and straightforward to implement, a MC surface may have discontinuity even though the volumetric data is continuous. This is because surface construction is more sensitive to image noise than the construction of volumetric data. To address this problem, we propose a surface construction algorithm which aggregates local surfaces constructed by the 3-D convex hull algorithm. Thus, the proposed method initially classifies local convexities from imperfect MC vertices based on sliced volumetric data. Experimental results show that continuous surfaces are obtained from imperfect silhouette images of both convex and nonconvex objects.

Triangular Mesh Generation of Octrees of Non-Convex 3D Objects

A general surface-generating algorithm, the marching cube, produces triangular meshes from octants where the vertices of octants are clearly classified into either inside or outside the object. However, the algorithm is ambiguous for octrees corresponding to non-convex objects generated using a shape from silhouette technique. This paper presents a methodology which involves Delaunay triangulation to generate surface meshes for such octrees. Since the general 3D Delaunay triangulation creates 3D convex hull which consists of tetrahedron meshes, we propose a method which applies the Delaunay algorithm locally in order to deal with non-convex objects. The proposed method first slices an octree and detects the clusters in each slice. All clusters between adjacent slices are linked based on a 3D probability density cube. The Delaunay algorithm is then applied to locally-linked clusters. Finally the accumulation of triangular meshes forms a final non-convex surface mesh

Evaluation of close-range stereo matching algorithms using stereoscopic measurements

The performance of binocular stereo reconstruction is highly dependent on the quality of the stereo matching result. In order to evaluate the performance of different stereo matchers, several quality metrics have been developed based on quantifying error statistics with respect to a set of independent measurements usually referred to as ground truth data. However, such data are frequently not available, particularly in practical applications or planetary data processing. To address this, we propose a ground truth independent evaluation protocol based on manual measurements. A stereo visualization tool has been specifically developed to evaluate the quality of the computed correspondences. We compare the quality of disparity maps calculated from three stereo matching algorithms, developed based on a variation of GOTCHA, which has been used in planetary robotic rover image reconstruction at UCL-MSSL (Otto and Chau, 1989). From our evaluation tests with the images pairs from Mars Exploration Rover (MER) Pancam and the field data collected in PRoViScout 2012, it has been found that all three processing pipelines used in our test (NASA-JPL, JR, UCL-MSSL) trade off matching accuracy and completeness differently. NASA-JPLs stereo pipeline produces the most accurate but less complete disparity map, while JRs pipeline performs best in terms of the reconstruction completeness.