Depth Measurement and 3D Reconstruction of the Scene from Multiple Image Sequence

Abstract

In the everyday life we continuously see many scenes and easily recognize them and perceive their 3D structure, this simple 3D scene reconstruction problem is a challenging task in the area of computer vision from the past few decades. In the present work, we explored few options to tackle this problem and try to establish the bound of these methods. In this project we, input the image sequence of the stationary scene from different views and get the depth of the image from the camera from which we ultimately reconstruct the 3D scene. We use epipolar geometry to find the fundamental matrix and match correspondences between images, we calibrate the camera intrinsic parameters and use them triangulate to get the depth of the object. An efficient dense matching has been implemented to increase the number of correspondences between a pair of image. This algorithm takes few correspondences the matches with highest some points of interest which have the highest gradient as the initial seed points and propagates the matches in the neighborhood of seed points from the most gradient pixels to less gradient ones. We also analyzed the increase in the number of the points by applying dense matching algorithm