The Research On Feature-based Method For 3D Information Reconstruction

Strcture From Motion (SFM) is the technology of obtaining the depth information of a scene from the image sequences. As one of the most challenging problem in computer vision area, it can be used to various applicatons.The problem our research faced is how to recover the 3D structure information by analysing and computing the 2D image sequences captured with an uncalibrated camera throughout different directions in a rigid static scene, assuming that the intrinsic parameter is invariable.This paper is is focused on the issues of the recovery of the 3D information of a realistic scene from image sequences and some critical issues related to this subject has been studied.The main work of the thesis is as follows:1. We analysed and compared some corner detection operators which are used frequently, extracted the corner automatically.A method which combines the corresponding of the corners and the fundamental matrix estimation was introduced to improve the corner matching and refine the estimation of fundamental matrix.2. We rectified the image pairs of which the fundamental matrix, has already been estimated by making all matching epopolar lines coincident and parallel with an image axis.3. A pixel-based dense matching method, using a maximum likelihood measurement, constrained by ordering constraint, uniquess constraint and disparity limit, is used to estimate the disparity of the image pair along the each epipolar line by dynamic programming and at last we get the dense depth map.4. We integrated all of the above algorithms and implemented a 3D information reconstruction system based on point feature.The experiment results demonstrate the effectiveness of the reconstruction system.