Font Size: a A A

Research And FPGA Implementation Of Binocular Matching Algorithm Based On SAD And Census Fusion

Posted on:2021-01-04Degree:MasterType:Thesis
Country:ChinaCandidate:N WangFull Text:PDF
GTID:2428330611970886Subject:Electronic and communication engineering
Abstract/Summary:PDF Full Text Request
With the development of video and image acquisition technology,binocular vision,as one of the popular research fields of computer vision technology,has been widely used in 3D reconstruction,VR/AR technology,robot navigation and other fields.Stereo matching technology is the core technology in binocular vision,but the existing matching algorithm has low matching accuracy,which directly affects the accuracy of 3D scene depth information recovery,so the research on stereo matching algorithm is of great significance.Based on the research,analysis and summary of the advantages and disadvantages of various matching methods,this thesis selects SAD(Sum of absolute differences)and Census as the basic matching cost calculation method,and introduces the image pixel gradient as the complementary matching cost to improve SAD based on the SAD algorithm.The matching cost solves the problem that the traditional SAD matching cost cannot reflect the pixel change information in the template;it is merged with the Census matching cost to improve the robustness to lighting changes;the image edge detection operator is used to add an adaptive selection matching template to solve the matching The effect of the template on the matching accuracy;the algorithm is tested using the Middlebury dataset.Because the improved algorithm increases the amount of calculation,the traditional software platform can not meet the real-time requirements.In order to accelerate the algorithm,this thesis selects FPGA(Field Programmable Gate Array)device to accelerate the algorithm,design image acquisition and preprocessing,and improved SAD-Census transformation,parallax calculation and other modules,cascaded to form a binocular stereo matching system after simulation verification.In the software testing stage,the algorithm of this thesis was tested using Visual Studio 2017.Under the test picture provided by the Middlebury data set,the matching accuracy reached 90.32%;in the hardware verification stage,the cascade binocular acquisition module,stereo matching module and VGA display module After testing the real experimental scene,the experimental results show that each module of the binocular stereo matching system works normally,and the processing time of the single-frame image is less than 0.017s,which provides a high-precision disparity map that meets the real-time requirements for the subsequent 3D reconstruction.
Keywords/Search Tags:Machine vision, Stereo matching, Adaptive template, FPGA
PDF Full Text Request
Related items