Large Scale 3D Scene Reconstruction Based On Array Cameras

Based on visual processing,the technology of acquiring three-dimensional information has become a very important research direction in computer vision field.At present,the commonly used methods about acquiring 3D scene information is mainly based on binocular stereo vision,structured light coding,TOF depth camera and so on.However,those methods that were mentioned above could not meet requirements of the multi-faced,high precision and fast 3D reconstruction.To the problems,under the large-scale scenes indoor,the paper construct 3D reconstruction system of camera array,based on binocular stereo vision,after stitching the points cloud that is reconstructed by binocular camera from different anglesm,get the panoramic multi-angle three dimensionnal reconstruction.This paper mainly includes array camera calibration,fast binocular stereo matching,three-dimensional reconstruction,the global matching of array camera system and the construction of relevant hardware and software.Firstly,based on binocular camera calibration,this paper designs a flexible array camera calibration method to realize the calibration of internal parameters and external pose parameters for the array cameras.The experimental results show that the calibration method can achieve high precise calibration,the accuracy of binocular calibration reaches 0.08 mm,and that of array camera calibration reaches 0.079 mm.Secondly,due to the lower speed and the larger amount of calculation of existing binocular matching algorithms,in this paper,the SIFT algorithm has been transplanted to the graphics processor unit(GPU)to speed up the extraction of feature points,and on this basis,a dense matching algorithm which is based on Bayesian model is used to achieve fast and dense binocular matching.Then,the external calibrat ion parameters of the camera array has been used to initialize the point clouds obtained from each angle of the views,furthermore,using ICP(Iterative Closest Point)algorithm to optimize the accuracy of point cloud stitching,and achieving the goal of high precision 3D scene reconstruction.Finally,the relevant platform of hardware and the software system of the camera array have been built.The hardware platform realizes the control of the camera array by signals,the synchronous trigger of camera sys tem and the transmission of the high bandwidth image data.In the software system,the camera array reconstruction algorithm has been integrated together to realize human-computer interaction and the efficient visualization.