Research And Application Of Patch Match Stereo Matching Algorithm Based On Adaptive Window

Vision is the main way for human to obtain external information.Through the visual system,we can quickly capture important information such as the color,shape and distance of external objects.With the gradual development of intelligence in all walks of life in modern society,computer vision related technologies have also entered all fields of our lives.However,most computer vision software is used to analyze and process two-dimensional images.With the improvement of practical application requirements,the defects of two-dimensional vision are becoming more and more obvious,However,3D vision related technology is increasingly demanded by the market because it can measure the features related to the shape of objects.As an important part of 3D vision,binocular stereo vision has the characteristics of low cost,wide application scenarios,simplicity and high accuracy,so it has been widely used and is also a hot research field in recent years.The main contents of this paper are as follows:(1)Firstly,this paper introduces the research status of stereo matching methods at home and abroad,then discusses according to the theory of binocular stereo matching,briefly introduces the theoretical basis of binocular stereo vision,then summarizes the principle of each step of binocular stereo matching,and introduces the commonly used algorithms in each step.Finally,among many stereo matching algorithms,the classical binocular stereo matching algorithm Patch Match Stereo is studied.(2)Aiming at the fact that the traditional Patch Match Stereo algorithm uses fixed size windows,this paper first experiments with windows of different sizes on Middlebury data set,makes intuitive and quantitative analysis on the results,and verifies the influence of window size on the effect of Patch Match Stereo algorithm.Then on this basis,this paper proposes an adaptive window method based on image gray information to replace the fixed window used in the original algorithm.This method measures the texture richness of the region according to the average value of pixel gray difference in the current window of a pixel in the image.For the region with rich texture information,the smaller window is directly used for matching to reduce the amount of calculation and improve the efficiency of the algorithm.For the region with less texture information,the window size is appropriately increased to obtain more reference pixels,so as to improve the accuracy of the algorithm.After that,this paper tests the improved algorithm on the cones,piano,Teddy and reindeer image sets of Middlebury data set,and comes to the conclusion that the overall efficiency of the algorithm is improved by 12.19% on average,and the error matching rate of parallax results is also reduced by 1.85% compared with the original algorithm.(3)For the Patch Match Stereo algorithm based on adaptive window,this paper also uses Azure Kinect DK depth camera to shoot multiple groups of image pairs in indoor and outdoor scenes.After resolution reduction,it uses Bouguet method for epipolar correction.Finally,it is brought into the algorithm to calculate the parallax.Good results are obtained,which verifies its feasibility and good generalization ability in the actual scene.In order to compare with the improved PMS algorithm,this paper uses the classical SGM algorithm to calculate the parallax of the same data.Finally,compared with the PMS algorithm based on adaptive window,the effect of this paper is better.