Research On Local Stereo Matching Algorithm For Indoor Complex Scenes

Stereo vision is one of the most important research directions in the field of photogrammetry and computer vision.By capturing two or more two-dimensional images of the same three-dimensional scene from different perspectives,stereo matching technology is used to find each pixel between each image.The matching point is calculated and the corresponding disparity value is calculated,and the depth information of the three-dimensional scene is restored according to the triangulation principle and the corresponding disparity information.Binocular stereo matching is the focus and difficulty of stereo vision.The accuracy of stereo matching will directly affect the results of 3D reconstruction.According to the optimization method of disparity computation,the stereo matching algorithm can be divided into global algorithm,semi-global algorithm and local algorithm.By constructing the energy function,the global algorithm transforms the problem of finding pixel matching points of two-dimensional images into an optimal solution problem of energy function minimization.The precision is high but the calculation is complex and the operation speed is slow.The semi-global algorithm is based on the idea of the global algorithm.The two-dimensional image energy function optimization problem is approximated by a one-dimensional image energy function in multiple directions.The dynamic programming idea is used to solve the optimal parallax in multiple directions.The global algorithm reduces time complexity and achieves good matching accuracy.The local stereo matching algorithm calculates the parallax by constructing a suitable support window and using the neighborhood pixel information in the window to aggregate.Compared with the global algorithm,the accuracy is low,but the calculation speed is fast,and the implementation is simple.However,with the advent of adaptive window and adaptive weighting,the accuracy of local algorithms can be compared with global algorithms,and the application of local algorithms is more and more extensive.The local stereo matching algorithm includes four steps of cost calculation,cost aggregation,disparity calculation and disparity refinement.In this paper,we will focus on the difficulty of matching complex scenes such as amplitude distortion,occlusion,depth discontinuity,weak texture and repeated texture in the indoor scene.The three steps of cost calculation,cost aggregation and disparity refinement of local algorithm are studied in detail.The main research contents of the thesis include the following aspects:(1)Aiming at the problem that the existing local stereo matching is not suitable for complex indoor scenes(such as amplitude distortion,object occlusion,weak texture,etc.),a local stereo matching algorithm based on improved cost calculation and adaptive guided filtering cost aggregation is proposed.The existing local stereo matching algorithm cost Filter algorithm is improved from two aspects of cost calculation and cost aggregation.Firstly,in the cost calculation stage,a weighted joint matching cost function combining the enhanced gradient information and the gradient-based Census transform is proposed,which enhances the robustness of the complex indoor scenes such as illumination changes and exposure levels.In the cost aggregation phase,it is difficult to solve the problem of large-area weak texture region matching for the traditional guided filter cost aggregation,and the traditional guided filter aggregation method is improved.(2)Aiming at the problem that the initial disparity map has a large number of mismatched points in occlusion,weak texture and disparity regions,a multi-level parallax refinement method is proposed.Firstly,the left and right consistency detection and classification are used to perform parallax interpolation on the abnormal pixel points with matching errors.For the problem that the matching accuracy of the inclined plane regions common in indoor scenes is not high,the method based on oblique plane smoothing optimization is used to improve the inclined plane area.The matching accuracy,and finally the sub-pixel enhancement algorithm to improve the overall parallax accuracy.In this paper,the proposed algorithm is validated by various data sets provided by the Middlebury stereo vision test platform.The experimental results show that the proposed algorithm can effectively improve the matching accuracy of complex indoor scenes with amplitude distortion,weak texture and repeated texture.