Catálogo de publicaciones - libros

Existing Photometric Stereo methods provide reasonable surface reconstructions unless the irradiance image is corrupted with noise and effects of digitisation. However, in real world situations the measured image is almost always corrupted, so an efficient method must be formulated to denoise the data. Once noise is added at the level of the images the noisy Photometric Stereo problem with a least squares estimate is transformed into a non-linear discrete optimization problem depending on a large number of parameters. One of the computationally feasible methods of performing this non-linear optimization is to use many smaller local optimizations to find a minimum (called 2D Leap-Frog). However, this process still takes a large amount of time using a single processor, and when realistic image resolutions are used this method becomes impractical. This paper presents a parallel implementation of the 2D Leap-Frog algorithm in order to provide an improvement in the time complexity. While the focus of this research is in the area of shape from shading, the iterative scheme for finding a local optimum for a large number of parameters can also be applied to any optimization problems in Computer Vision. The results presented herein support the hypothesis that a high speed up and high efficiency can be achieved using a parallel method in a distributed shared memory environment.

A specific diagram language, called the floor-layout language, is considered. It allows to edit floor-layouts maintaining specified restrictions. An internal diagram representation in the form of hypergraphs needed for syntactic analysis is described. Some examples of designing floor-layouts are presented.

This paper describes a method for modeling wide area outdoor environments by integrating omnidirectional range and color images. The proposed method efficiently acquires range and color data of outdoor environments by using omnidirectional laser rangefinder and omnidirectional multi-camera system (OMS). In this paper, we also give experimental results of reconstructing our campus from data acquired at 50 points.

Exploiting the simplification hierarchy and hypertriangulation model, a unified framework (ADSIMP) has been proposed for user-controlled creation of multiresolution meshes. For this system, here we present a new set of compact data structures for holding the multiresolution model, which is not only memory efficient but also supports the creation of hypertriangulation model simultaneous with bottom-up vertex hierarchy generation. The new proposed construction algorithm, based on multiple choice optimization simplification technique, is faster and makes available ADSIMP for real time applications. Comparison with related work shows that the new system provides combined environment at reduced overhead of memory space usage and faster running

We present a 3D object retrieval system in which a surfel-based model serves as a query and similar objects are retrieved from a collection of surfel-based models. The system comprises a surfelization technique for converting polygonal mesh models into a corresponding surfel-based representation, pose normalization of surfel-based models, a depth buffer-based method for describing shape of surfel-based models, and a search engine. Our surfelization technique consists of an enhanced triangle rasterization procedure adapted on the original triangulated model geometric features. Surfel-based representation is normalized by applying a modification of the Principle Component Analysis to the set of surfels. The 3D-shape descriptor is extracted from the canonical coordinate frame of the surfel-based model using the orthographic depth images. Each image is used as the input for the 2D Fast Fourier Transform. Appropriate magnitudes of the obtained coefficients are used as components of the feature vector. The resulting feature vector possesses an embedded multi-resolution representation, is invariant with respect to similarity transforms and robust with respect to outliers. The retrieval effectiveness of the presented method for characterizing shape of surfel-based models is evaluated using precision-recall diagrams.

It is presented a formal approach to the description of irregular geometrical forms by hyper-relations. This approach is based on an extension of the method of composite objects description based on the algebra of relations. It is shown that in some applications hyper-relations suit to the description of irregular geometrical forms better than other conventional tools. The general concept is illustrated by an example of irregular forms description by compositions of Freeman chains.

In this research we describe 3D data processing operations which were applied to 3D laser scans of human heads before 3D eigenfaces could be obtained both in shape space and color space. Modified median filtering was used to remove laser scan spikes. 3D scans registration exploits mapping of 3D mesh to the reference 3D mesh. The mesh is based on Facial Action Points (FAP) of MPEG-4 and the mapping itself on barycentric local coordinates in each triangle. 3D eigenfaces are computed by Singular Value Approximation (SVA) applied directly to 3D scans avoiding computation of prohibitively large covariance matrix.

The wireless capsule endoscopy (WCE) imaging technique provides detailed images of the gastrointestinal (GI) tract, in particular the small intestine, not feasible with earlier techniques. We present a model of deformable rings (MDR) for aiding in the interpretation of WCE videos. The model flexibly matches consecutive video frames with regard to displacement of distinctive portions of the GI tract’s tube-like surface. It creates a map, 2D representation of the surface, and estimates relative velocity of a capsule endoscope as it traverses the GI tract. The map can be glanced through for rapid identification of the possible abnormal areas for further detailed endoscopic video investigation. This method significantly reduces the total time spent in interpretation of the WCE videos.

The stereovision sensor described in this article has been developed during a research project called RaViOLi, for “Radar and Vision Orientable, Lidar”. The main outcome of this project is the improvement of driving safety thanks to the analysis of redundant data coming from several cooperative sensors installed on an autonomous vehicle. One is a high precision stereovision sensor whose field of view can be oriented toward a region of interest of the 3D scene, like the road in front of the vehicle at long distance. The sensor is composed of a single camera, of two lateral mirrors, and of a prism rotating about its edge. The mirrors project the left and right images of the stereo pair onto both halves of the imaging surface of the camera, yielding the equivalent of two virtual cameras with parallel axes. Rotating the prism changes the orientation of both optical axes while keeping them parallel.

This paper proposes a new method for detecting vehicles from an image. The method consists in three stages of segmentation, extraction of candidate window corresponding to a vehicle and detection of a vehicle. From the experimental results using 121 real images of road scenes, it is found that the proposed method can successfully detect vehicles for 120 images among 121 images.