Research On Steady Matching Algorithm To Target Based On Convariance Matix

Object matching based on the covariance matrix is one of the important branches ofimage matching, and the image matching is an important research area and technique ofthe fields of the computer vision and image processing. The core idea of the imagematching algorithm is that, firstly selecting the common features of both template imagesand images ready for matching, and then calculating the existed features with the methodsof correlation algorithms and the search strategies. Finally we will get the best matchingposition or corresponding points. Usually, the ideal matching algorithms can meet most ofthe requirements of high matching probability, small error, high speed and strong robust,this requires that algorithms with the ability to generate reasonable searching strategiesand to select appropriate calculating measurements are chosen.In this paper, we mainly study the implementation and application of the imagematching algorithm based on the covariance matrix. First of all, we make careful researchand summary of the existing image matching algorithms, and analyze the advantages anddisadvantages of them. At the same time, some traditional matching algorithms and imagefeatures are analyzed, and the experimental results are followed.Then we give a detailed description of the processes and results of matching based onthe covariance matrix, and compare it with other image matching algorithms. We focus onthe three steps of the matching: First, extracting the common image characteristicsbetween the two matching images, such as: image gradient amplitude and anglecharacteristics, morphological characters, the information entropy and spatial informationcharacteristics, etc; Second, constructing covariance matrix, the covariance matrix itselfhas many features which can be used to image matching, such as: overcome outside noiseor illumination of the interference, good robustness to the spin of the target, reducing theamount of calculation and so on, therefore, the article selects seven target features toconstruct the covariance matrix, thus it fulfills the task of accurately target matching;Third, selecting the measure of correlation algorithms, in this paper we convert thematching between the two images to calculate the similarity between two low dimensional matrixes, thus, we calculate the Euclidean distance to realize the matching. Finally, wepresent the experimental results in all kinds of conditions, these results prove that theimproved algorithm is better than the existing matching algorithms, and is of greatpractical value.Finally, we provided the application examples using the covariance matrix in targettracking. In the first part, we make an explanation of the singular value function ofreducing the noise, and prove the advantages of it. Then in the second part, we makeresearch to the target tracking based on the covariance matrix, and show some resultimages of tracking. At the same time, we make some experiments of target tracking whosepictures are impacted by the adding noise or the natural noise. The results proved theexcellent characteristics of the covariance matrix. In the third part, we apply thecovariance matrix to the particle filter tracking, and complete the mixture of thecovariance matrix and the particle filter tracking. In the end, we show some trackingrenderings based on the improved particle filter algorithm. The tracking experiment resultsabove prove that the covariance matrix possesses the application advantages in thetracking system.