Elsayed E. Hemayed

Neurocalibration: a neural network that can tell camera calibration parameters

Camera calibration is a primary crucial step in many computer vision tasks. We present a new neural approach for camera calibration. Unlike some existing neural approaches, our calibrating network can tell the perspective-projection-transformation matrix between the world 3D points and the corresponding 2D image pixels. Starting from random initial weights, the net can specify the camera model parameters satisfying the orthogonality constraints on the rotational transformation. The neurocalibration technique is shown to solve four different types of calibration problems that are found in computer vision applications. Moreover, it can be extended to the more difficult problem of calibrating cameras with automated active lenses. The validity and performance of our technique are tested with both synthetic data under different noise conditions and with real images. Experiments have shown the accuracy and the efficiency of our neurocalibration technique.

Novel surface registration using the grid closest point (GCP) transform

A novel approach has been developed for fast registration of two sets of 3-D curves or surfaces. The technique is an extension of Besl and Mackays (1992) iterative closest point (ICP) algorithm. This technique solves the computational complexity associated with the ICP algorithm by applying a novel grid closest point (GCP) transform and a genetic algorithm to minimize the cost function. A detailed description of the algorithm is presented along with a comparison of its performance versus several registration techniques. Two applications are presented in this paper. In the first, the algorithm is used to register 2-D head contours extracted from CT/MRI data to correct for possible mis-alignment caused by motion artifacts during scanning. In the second, the algorithm is used to register 3-D segments of the human jaw obtained using the shape from shading technique. Registration using the GCP/GA technique is found to be significantly faster and of comparable accuracy than two popular techniques in the computer vision and medical imaging literature.

3D reconstruction of the human jaw from a sequence of images

A novel approach is proposed to obtain a record of the patients occlusion using computer vision. Data acquisition is obtained using intra-oral video cameras. The technique utilizes shape from shading to extract 3D information from 2D views of the jaw, and a novel technique for 3D data registration using genetic algorithms. The resulting 3D model can be used for diagnosis, treatment planning, and implant purposes. The overall purpose of this research is to develop a model-based vision system for orthodontics to replace traditional approaches. This system will be flexible, accurate, and will reduce the cost of orthodontic treatments.

Sensor planning for a trinocular active vision system

We present an algorithm to solve the sensor planning problem for a trinocular, active vision system. This algorithm uses an iterative optimization method to first solve for the translation between the three cameras and then uses this result to solve for parameters such as pan, tilt angles of the cameras and zoom setting.

CardEye: a 3D trinocular active vision system

CardEye is an experimental, trinocular, 3D active vision system. Our goal is to create a flexible, precise tool for active vision research. The system uses an agile trinocular vision head mounted on a robotic arm. It has five degrees of freedom: pan, tilt, roll, vergence and variable baseline in addition to the automated zoom and focus of the lenses. It utilizes an active lighting device to assist in the surface reconstruction process. In this paper we describe the architecture of the system together with its functionality.

A neural approach for single- and multi-image camera calibration

This paper presents the neurocalibration approach as a new neural-based solution for the problem of camera calibration. Unlike some existing neural approaches, our calibrating network can match the perspective-projection-transformation matrix between the world 3D points and the corresponding 2D image pixels. Starting from random initial weights, the net can specify the camera model parameters satisfying the orthogonality constraints on the rotational transformation. In order to improve the accuracy of calibration results, the paper demonstrates the application of the neurocalibration technique to multi-image camera calibration. In such a case, many images are taken by the same camera but from different (rotated and/or translated) positions. Experiments have shown the accuracy and the efficiency of our neurocalibration technique.

A geometrical-based trinocular vision system for edges reconstruction

We present a trinocular stereo vision technique that can be used to build a fast, robust description of indoor objects. The proposed technique utilizes the geometrical constraints of the trinocular vision system and considers the uncertainty of image measurements. The technique has been successfully applied to several indoor scenes. The reliability and speed of the proposed system make it particularly suitable for real-time applications such as robot navigation.

Investigation of stereo-based 3D surface reconstruction

This article presents an investigation study of stereo-based 3D surface reconstruction algorithms by providing an overview of the different approaches that have been investigated in the stereo literature during the last decade. This study considers only the two-views plain stereo algorithms and provides another classification for the stereo approaches based on the features used in the stereo literature. In addition, the article provides full details of two different stereo algorithms that give an idea of how stereo works.

Target recognition via 3D object reconstruction from image sequence and contour matching

Abstract This paper proposes an automatic target recognition (ATR) system based on the three-dimensional (3D) reconstruction of the target from an image sequence. The main contribution of this work is twofold: (1) we present a modified voxel coloring reconstruction algorithm and (2) we employ the 3D reconstructed target model to generate the front and side target templates at zero depression angle to be used in the target recognition process. Target recognition is performed by matching the generated templates to a library using a subpixel contour matching algorithm. Experimental results on simulated scenes show the accuracy of the approach presented in this paper.

Slicing, Fitting and Linking (SFL): A Modular Triangulation Approach

Slicing-fitting-linking (SFL) is a fast triangulation technique that guarantees building a closed mesh with consistent normals. The proposed technique can be used with different surface reconstruction cues such as laser scanner, stereo, SFS, and CT/MRI. The output SFL can be in the form of STL files that are suitable for most rapid prototyping machines. The technique has three tasks. The first task is to slice the 3D data points into 2D cross sections parallel to each other. The second task is to fit a curve to the data points of each cross section. The third task is to link the fitted curves to form the mesh. A detailed description of the algorithm is presented in this paper.