Three-dimensional object reconstruction using structured light and two two-dimensional images.

Abstract

For machine vision and for graphical representation of real objects in a computer environment, the three-dimensional (3D) reconstruction of the image of a real object has become a key technique. A few methods such as range finding, which use laser scanners, computer tomography, based on CTR or MRI machines, computational stereo, etc. were developed. Computational stereo is broadly defined as the recovery of 3D characteristics of a scene from a series of images obtained from different points in the three dimensional (Euclidean) space. In this thesis, a new algorithm and system is introduced and developed for the 3D reconstruction of the images of real objects from two 2D images required with two cameras. The algorithm is based on a new matching method, a new procedure for the determination of the fundamental matrix used in stereo vision, and a new technique for stereo fusion. The novelty of the matching procedure, and of the determination of the camera alignment, consists of the projection of a structured light pattern on the real object, the pattern being created using a pseudo-random encoded mesh (PRBA) (34). The novelty of the stereo fusion algorithm consists of the application of the dynamic programming principle (DP) (5) (27) using a cost function which contains the information obtained from the list of matched points. It is also proposed to use an autoregressive (AR) modeling technique for calculating the stereo disparity of each pixel of the two images. The autoregressive filter helps the DP part of the algorithm to calculate the disparity of the pixels when the above pixels are occluded. The above proposed methods offer three distinctive advantages over a conventional stereo system: (1) It easily generates a list of matching points. (2) It adds structure to an object without textures. (3) It is less computational intensive.