Research On Depth Imaging Algorithm Based On Binocular Vision

The depth imaging technology is an important method for 3D scene information acquisition.It is also an important field of computer vision.The depth imaging technology is widely used for human-computer interaction,robot vison,3D reconstruction,non-contact measurement and so on.The depth imaging method based on binocular vision is one of the methods for obtaining depth image.Compared with other methods,it has the advantages of long service life,easy realization and low cost.Nevertheless,because of the limited calibration accuracy,the disparity exist not only in the latitudinal neighborhood but also in the longitudinal neighborhood.However,the traditional stereo matching algorithm only considers the influence of the latitudinal neighborhood.Thus,without considering the 2D disparities,the accuracy of the stereo algorithm and the calculated depth is poor.To solve this problem,with the consideration of the 2D disparities,this paper proposed the LL-swap-move stereo matching algorithm based on the alpha-beta swap algorithm.Firstly,in order to correspond to the 2D disparities,the 2D labels are designed.Then a network graph is built according to the 2D labels.In addtion,the weights of the edges in the network are improved with the gradient and smooth information.After that,max-flow algorithm can be used to obtain the min-cut of the network.Finally,the label assignment is completed with the improved upated strategy and the dispairy map is generated.At the same time,the performance of the proposed algorithm is proved from the perspective of maximum posterior probablity estimation(MAP).In this paper,to realize the depth imaging algorithm based on binocular vison,the LL-swap-move stereo matching algorithm is combined with camera calibration,disparity map post-processing and depth imaging.To verify the effectiveness of the the proposed algorithm,the actual data experiment and the open dataset test are conducted.In the comparative experiment with 2D disparities,the minimum mean square error of the alpha-beta swap algorithm are 3.30px in the latitudinal neighborhoods and 19.23px in the longitudinal neighborhoods.However,the errors of LL-swap-move algorithm are only 0.41px in the latitudinal neighborhoods and 0.84px in the longitudinal neighborhoods.In addition,the results of the proposed algorithm are similar to the results of traditional algorithm on the open dataset.The experiments show that LL-swap-move algorithm has better optimization performance.The proposed algorithm can retain contour information better in strong texture regions.Moreover,in weak texture regions,the results of this algorithm are smoother.It is verified that LL-swap-move algorithm is suitable for binocular depth imaging with 2D disparities.