3D Reconstruction Algorithm Study Of Large View Based On EPI Sequence Images

The three-dimensional reconstruction and image stitching based on depth data have been research hotspots in the field of computer vision all the time,so how to obtain the accurate depth value has become the focus of people’s research.In recent years,the depth recovery based on the light field sequence image has become a relatively new development direction.Because it no longer involves the camera calibration,the basic theoretical knowledge is no longer so complicated.But so far the researche achievements in this respect are relatively small.It has not yet formed a set of complete research system that the depth reconstruction based on the light field and large view stitching and three-dimensional modeling based on the depth information.In this paper,we conducted an in-depth study on the depth acquisition based on the sequence image,the large view stitching algorithm and three-dimensional reconstruction algorithm based on depth data.The specific research content includes the following aspects:(1)In this paper,an adaptive depth reconstruction algorithm based on EPI image is proposed.We can obtain the image sequence by using the built image acquisition system and generate the EPI image by preprocessing the image sequence.The EPI image is a polar plane image,which is obtained by "slicing" the sequence of the light field image.Due to the linearity of the EPI image,the algorithm divides the straight points into inside class points and outside class points according to the adaptive framework,making the energy function calculation only occurs at the interior point,thus obtaining a precise edge depth.Finally,algorithm diffuses the non-edge region,and obtains a priori value of depth by using the bilateral model,and then diffuses the depth to the whole area through the strategy of the winner take all,and finally the dense depth image will be obtained.(2)A large view image stitching algorithm based on the depth data is proposed.Compared with the color image,the depth image has fewer feature points and the feature is difficult to extract.For this problem,paper proposes a feature extraction method based on edge information.That is,after extracting the edge,the corner feature extraction is carried out on the basis of the edge image.Due to the particularity of the photographing,a matching method based on the similar pixel column is proposed,that is,the distance between the adjacent images is obtained by using the parallax calculation principle of the adjacent pixels on the basis of the corner feature information.Finally,we use the improved weighted average fusion algorithm to deal with the image matching area,and finally get a large view of seamless high resolution depth image.(3)The three-dimensional model reconstruction algorithm based on simple grid extraction is proposed.Firstly,according to the principle of spatial three-dimensional relational reasoning,the three-dimensional coordinates of the pixel information in the depth image are calculated.Then,according to the point spacing and the normal vector angle,we can judge the triangle connection relationship between the point and point and obtain the dense triangulation.Finally,using depth and color image of the large view as input,we can reconstruct the 3D model of scene with wide viewing and angle texture.