3D Shape Extraction from Uncalibrated Environments and Video Camera A. S. T. HUSSAIN (1,2) N.E. BERRACHED (1) A. E. MURAD (2) TAYEB BASTA (2) (1) University of Science and Technology of Oran (USTO), Oran 31000, Algeria. (2) University of Sharjah, Sharjah, UAE. emails: asthussain@yahoo.com Or asthussain@sharjah.ac.ae Tele. +971-50-6676753 & Fax. +971-6-8838555 ABSTRACT In this paper, we have described an approach for a 3D scene reconstruction using 2 randomly selected adjacent colored video frames. We have used a single uncalibrated video camera to take a record for uncalibrated environment. The Selection of 2 frames based on the maximum homogeneity of points on these frames are favorable; could be any two adjacent frames. The use of Harris technique were very useful to find the edges and corners on each selected image (frame), then the use of the autocorrelation function based on Gaussian’s function been used to find the corresponding matched points. Then the correlated matched pair points are found on both images, and by calculating the gradient of the correlated paired points on both images represents approximately the Z direction (calculating dzdx and dzdy). This is yielding that each point on each image (frame) can be represented in a 3D coordinates which yields to 3D shape estimation, which is achieved by the RANSAC function. This method has got an errors, because of its dependence on probability. The main advantages of the proposed approach is applicable for indoor and outdoor application. This technique is suiting the natural real world applications. The proposed method is illustrated on a set of examples of an indoor captured colored video records, the selected frames are selected randomly from the video records. Key Words : 3D Reconstruction Autocorrelation Gaussian RANSAC 1. Introduction Reconstructing the scene from image sequences captured by moving cameras with varying intrinsic parameters is one of the major achievements of computer vision research in recent years. The reconstruction process is a set of consecutive tasks to be accomplished. First task consists of extraction of point landmarks, lines/curves, and surfaces/regions in 2D. Secondly, regarding the geometry used; two frames means two-views, relevant methods for determining the correspondence between features are used. Then, appropriate transformation functions for image registration are applied. Finally, determination of the reliability, accuracy, and speed of applied image registration methods are verified and compared to their counterpart available in the literature. Here, we are presenting a light survey in which the problem of 3D extraction from a frame sequences is tackled. After that, we investigate the literature for the use of two- views to extract 3D structure. The basic geometrical and algebraic notions that make the foundation for such extractions follow the latter investigation. After that, we shade some light on different image registration techniques and finally we quote some words about camera self- calibration (Uncalibrated) techniques as we think that it can be exploited somehow to reach our objective in many different ways [ 1, 2, 3, 4 & 10].