B-rep object description from multiple range views

Abstract

We present a system which computes an integrated description of an object from multiple range images. The object description is in the form of B-rep (boundary representation), which has not been achieved by the computer vision community. To do so, we emphasize the inherent difficulties and ambiguities in the low to mid level vision, and present novel techniques of resolving them. In this system, each view of the object is represented as an attributed graph, where nodes correspond to the surfaces (vertices) and links represent the relationship between surfaces. The main issue in surface extraction is contour closure, which is formulated as a dynamic network. The underlying principle for this network is weak smoothness and geometric cohesion, and is modeled as the interaction between long and short term variables. Long term variables represent the initial boundary grouping computed from the low level surface features, and short term variables represent the competing hypotheses that cooperate with the long term variables.The matching problem involves matching visible surfaces and vertices, and provides the necessary basis for volumetric reconstruction from multiple views. The matching strategy is a two step process, where in each step uses the Hopfield network. At each step, we specify a set of local, adjacency and global constraints, and define an appropriate energy function to be minimized. At the first level of this hierarchy, surface patches are matched and the rigidity transformation is computed. At the second level, the mapping is refined by matching the corresponding vertices, and the transformation is verified.The multiple-view reconstruction consists of two steps. First, we build a composite graph that contains the bounding surfaces and their corresponding attributes, and then intersect these surfaces so that the edges and vertices corresponding to the B-rep description are identified. We present results on objects with planar, as well as quadratically-curved, surfaces.

Edge Detection in Range Images Based on Scan Line Approximation

“…For this purpose, however, some contour closure algorithm [12,13] is needed because of the inherent gaps in an edge map. The usefulness of the edge detection method proposed in this paper has been tested by using a simple edge grouping technique [13].…”

“…Edge detection has the potential of a complete segmentation. For this purpose algorithms [12,13] have been developed to effectively close gaps in edge maps of range images. Alternatively, we can make use of edge detection to support region-based segmentation.…”

Saliency Sequential Surface Organization for Free-Form Object Recognition

“…Parvin and Medioni [61] proposed a system for object description and matching from multiple range views. Notable in this work is the labeling and grouping of 3D edges (creases and limbs) and vertex description for model-building.…”

Rapid and non-invasive surface crack detection for pressed-panel products based on online image processing

“…Jiang and Bunke 22 developed an edge detection algorithm based on a scan line approximation technique in order to overcome the limitations of existing algorithms on object perception. 23 Their developed algorithm generates the shape of objects based on the strength of edge lines. Abdel-Qader et al 24 compared the performances of four crack detection techniques, including fast Haar transformation (FHT), fast Fourier transformation (FFT), Sobel filter, and Canny filter.…”