Modelling of a Three-dimensional Image from 2D Stereo Pairs

N. Vira (USA)


Modeling, visualization, and image processing


In recent years, there has been a growing interest in the visualization of range images to map scenes arising in many of the military, industrial, and medical applications. Computational stereopsis, where disparity is computed from two or more cameras with displaced image planes and then reconstructing three-dimensional scenes via stereo triangulation method, is a powerful means for the passive acquisition of range information. This paper describes a method (coded by Sun Microsystems's Java programming) to create three dimensional image geometry from a pair of digital images. A general purpose interactive JView program engineered by Air Force Research Laboratory, Rome, NY was utilized to recreate 3D-textured image model. Though several procedures have been proposed in literature on how to generate 3D models from 2D stereo pairs by recovering the depth, they are limited in predicting an accurate pixel matching between image correspondences. Furthermore, the majority of papers lack depicting the reconstructed 3D-image geometry to show how well their adopted techniques have been able to recover the scene depth at every pixel. We have employed a region-based block matching methodology for color image analysis. The reconstructed 3D texture mapped image model can be viewed and maneuvered on a computer screen by moving up and down, rotating around, and zooming in and out with the help of mouse buttons. This interactive viewing capability is an aid to the visualization process and facilitates algorithm designers to test step-by-step their codes and make appropriate judgments as to the accuracy of the underlining reconstruction techniques (sort of a visual debugger). For the color images under investigation, we were able to reconstruct the three-dimensional building cube within the accuracy of 2.4% error in relation to the camera viewing height

Important Links:

Go Back