Research On Stereo Matching Algorithm Based On Census Transform And FPGA Implementation

With the continuous development of computer technology,binocular stereo vision technology has become more and more popular among scholars.Because binocular stereo vision is a non-contact measurement,it is widely used in aerospace,robot navigation,security monitoring and other fields.Stereo matching is the most important part of binocular stereo vision,and the matching result directly affects the effect of binocular stereo vision.The essence of stereo matching is the process of obtaining disparity map through stereo matching algorithm.The pros and cons of disparity map directly depend on the precision of stereo matching algorithm,which is of great significance for the research of stereo matching algorithm.This paper first analyzes and compares the advantages and disadvantages of various stereo matching algorithms.Under the premise of ensuring matching effect and matching speed,this paper finally chooses Census local stereo matching algorithm as the main research algorithm of this paper.Since the traditional Census stereo matching algorithm adopts a fixed size and a single transform window in the Census transform stage,it is impossible to dynamically select a suitable transform window,resulting in poor matching in areas such as discontinuous depth and weak texture of the image.Aiming at this problem,this paper adopts the Census stereo matching algorithm of adaptive window;and in the threshold selection,this paper adopts the adaptive threshold method to dynamically determine the size of the transform window in the Census transform stage.Finally,the algorithm of this paper is tested on the MATLAB software platform using the test images provided on the Middlebury stereo matching algorithm evaluation website.Experimental results show that stereo matching effect is remarkable in areas of image that are not deeply continuous,weakly textured.This paper improves the Census stereo matching algorithm and improves the matching accuracy of the algorithm.When the threshold is selected,the adaptive threshold of each pixel increases the complexity and computation of the algorithm and reduces the matching speed of the algorithm.Therefore,in order to improve the practicability of the algorithm,this paper uses FPGA to accelerate the hardware of this algorithm.Firstly,the overall framework and module division of hardware implementation are introduced.Secondly,Modelsim is used to simulate the main modules of hardware design,and the waveform diagram of function simulation is given.Finally,the hardware design of this paper is carried out on FPGA development board.The board level verification is performed,and the function simulation waveform diagram is compared with the board level verification waveform diagram,and the results are consistent.The experimental results show that the final processing speed can reach 32fps for image pairs with resolution.