Low-textured Scene 3D Reconstruction Based On Binocular Stereo Vision

The development of Internet technology has greatly facilitated our life,however,as an information carrier,image do not meet the needs of us.Recently,a mode with 3D model as a carrier and roaming as the information presentation way,becomes more and more popular,which has more stronger expression ability and receives more recognition.There are many methods for 3D modeling and the method based on binocular vision can provide great convenience and meet the requirements of general accuracy.The result of binocular 3D reconstruction is greatly influenced by the scene.For the reconstruction of low-textured scene,the general matching algorithm is unable to recover the depth information.To solve this problem,this paper proposes a hybrid stereo matching algorithm based on recursive perceptual filtering and image segmentation and use this algorithm to establish the 3D model of low-textured scene.For the needs of matching,we have to make a stereo calibration of binocular camera in advance,get the camera's internal and external parameters,and use these parameters to calibrate the binocular image to reduce the complexity of the matching point search.The matching algorithm is the key of our paper,we use recursive filtering to calculate the initial disparity map and reassigned the disparity of un-texture regions which is obtained by image segmentation algorithm and the related criteria.Before obtaining point cloud using triangulation principle,we carry on the sub-pixel disparity map to prevent point cloud fault.After obtaining the point cloud,related filter and the triangle mesh used to the point cloud and finally forms the three-dimensional model.The disparity map and 3D model are analyzed and compared from the qualitative and quantitative point of view.We find that the algorithm we proposed can effectively recover the depth information of the low-textured region,reduce the possibility of failure matching and improve the matching efficiency of the whole scene.