Research On Fast Parallel Image Matching Algorithm Based On Feature Points

With the rapid development of computer technology and information science,image matching techniques are widely used, and its performance requirements are alsorising. At present, the problems of image matching techniques are mainly in thefollowing two aspects: the calculation speed is less than real-time requirements; therobustness is not high enough under block, large-scale changes, illumination changes,rotation changes, blur changes and so on. So we launched the research on the imagematching technology, to improve the speed and robustness of the algorithm. The maincontents can be summarized as the following:①This paper introduces the present state of the research in image matchingtechnology, which is the theoretical basis for the below algorithm. We introduced thebasic process of image matching techniques, outlined the approach of the imagepre-processing of grayscale and the theoretical basis of mismatch constraint ofRANSAC algorithm, and listed the experimental images test library in the field ofimage matching, as well as experimental evaluation criteria. We also studied on cameracalibration method in the theoretical preparation stage, and selected the standard gridmethod to do the camera calibration experiments. The above work builds a solid basefor the research of fast image matching technology.②We realized an image matching algorithm based on DAISY descriptor. Based onthe FAST corner detector and BBF approximate search algorithm, we used the latestDAISY descriptor to achieve a fast and robust image matching algorithm. A lot ofexperiments were done, respectively, to verify the excellent performance of the FASTalgorithm, the DAISY algorithm. And our new matching algorithm is compared with theSIFT algorithm which is currently the most popular experiments, from a subjectiveevaluation and objective evaluation, to prove its high efficiency and practicality.③For the challenge of the real-time requirements in the practical application ofimage matching algorithm, this paper presents a fast parallel matching algorithm basedon feature points. The algorithm uses a two-step coarse to fine matching method, andtakes advantage of the computer dual-core technology to create the parallel thread ofmapping, to achieve parallel two-step matching algorithm based on feature points.Experimental results show that our algorithm can be successfully applied to real-timeenclosures tracking, computing speed and matching performance can meet the real-timerequirements, and under large-scale view changes, block, blur changes and rotation changes, our method can run robustly. Our research provides a new way of thinkingabout the image matching.