Research On SoPC Design Technology Of Image Matching System Based On SIFT-RANSAC Algorithm

With the enhancement of the computing performance of chip-level integrated circuits,as well as the improvements of systematic utilization of calculating resources,the algorithms whose realization on the software terminal costs large amounts of resources and time in the past now have greater feasibility in the realization on the hardware platform.And the algorithms in the field of image matching are among them.Because of their wide application in many scenes,such as satellite remote sensing,medical imaging,environmental detection,image stitch and robot localization and navigation,Image matching algorithms are research hotpots in the field of computer vision all the time.Some image matching algorithms which are commonly used are based on the grayscale of images,while others which include SIFT algorithm are based on the characteristic of images.In the image matching optimization system of this article,the keypoints' descriptors of the images will be generated after processing by SIFT algorithm.Then the matching pairs of keypoints will be calculated with the application of Euclidean distance matching algorithm.These are preliminary matching results.Due to natural characteristics of numerical calculation,some keypoints whose locations are not in the corresponding matching areas will be characterized by descriptors which have similar structures.Then some matching pairs which are incorrect actually will be generated by numerical matching.At this moment,it is required that the model of dataset be established.The model will indicate the mapping relations of the dataset.So it will select the correct matching pairs and improve the matching results.Therefore,RANSAC algorithm is applied in the system to optimize the image matching results.It will compute the mathematic model that reflects the mapping relations of the dataset,which includes erroneous matching pairs,then correct matching results will be obtained.On this basis,it will be beneficial to the application of the system in many scenes,such as the objects' localization,navigation and tracking.Moreover,the Homography Matrix which characterizes the Projection Transformation relation of the two images in the space will be obtained in the process of selecting the inliers.By applying the Homography Matrix to transform the source image,the matching area in the matching image will be identified,which will further enhance the applicability of the system.The embedded systems nowadays are in common use,which expands the scenes of the systems' practical application.Therefore,it is of great significance to transfer the image matching optimization system's operation from traditional software terminal to the SoPC platform based on FPGA.When operated on the software platform,it is unavoidable that the operation of the RANSAC algorithm will rely on some internal function libraries.So it is required that the operation detail be understood deeply and fully when transferring the algorithm to hardware platform.Moreover,in order to improve the real-time ability of the system,when operated on the SoPC platform based on FPGA,the circuitry structure of the realization of the algorithm needs to be designed specially.So that it can make full use of the characterization of the hardware platform to realize the parallel pipeline calculation,which will greatly increase the operation speed of the algorithm and improve the system's performance.The research on SoPC design technology of image matching optimization system based on RANSAC algorithm in this article realizes the operation of the RANSAC algorithm on the XCZ7020 platform that belongs to the serial Xilinx Zynq-7000 platforms by using the hardware and software co-design method and adapting to the features of the design of hardware logics.Then it will be integrated with SIFT Image-matching algorithm and Euclidean distance matching algorithm,This has greatly promoted the completeness of the image matching optimization system based on the SoPC platform.