Enhancement of 3D reconstruction process in terms of beautification and efficiency using geometric constraints

Abstract

In this paper, we present a 3D reconstruction approach from uncalibrated views using geometric constraints. Basically speaking, we used bundle adjustment based on Levenberg-Marquardt optimization with the aim to estimate our 3D scene. In fact, it is different to the classic case. We integrate a pose estimation algorithm in 3D reconstruction process. As it is known, Levenberg-Marquardt algorithm presents low convergence rate 0% if initial values are wrong. The use of pose estimation previously cited can improve convergence but, it is still not satisfactory for users. So, using geometric constraints present a good solution. It brings us many advantages; it helps us to reduce estimated parameters number and stabilizes good quality for 3D results. In fact, we should recall that we use uncalibrated views, so we don't have any prior information about our 3D scene to achieve 3D reconstruction with no pertinent initial values used in Levenberg-Marquardt algorithm. In this present work, we try as much as possible through a comparative analysis to proof the importance of geometric constraints use in 3D reconstruction in terms of results reliability, process speed and convergence rate. Several data will be used in the purpose to demonstrate the efficiency of our present approach using geometric constraints.