Research On Feature Matching Algorithm And Its Application In Target Tracking

Image matching technology has always been one of the hot topics in the field of computer vision.It has wide application prospects in many aspects such as intelligent surveillance,medical treatment,cultural relic restoration,image retrieval,image mosaic and so on.In intelligent monitoring,the most popular way is real-time tracking of targets.In real environment,the tracking effect of target can't achieve ideal state due to various complex effects of external and target itself.such as the change of the self shape of the non rigid object,the changes of the camera's angle of view,the change of the outside light and the occlusion of other targets.For these complex situations,there is not a tracking method that can solve all the situations.Compared with other methods,the feature matching target tracking method show good robustness based on can solve the occlusion,illumination changes,noise and non rigid target deformation to a certain extent,so the in-depth study and analysis of feature matching method,combined with feature matching algorithm for target tracking method has certain practical significance.Image matching technology is one of the core technologies of intelligent monitoring,and the development of image matching technology is of great significance to the development of video surveillance.At present,SIFT algorithm and ORB algorithm get the favor of researchers,but because SIFT algorithm takes a long time to detect the global feature points,which causes the algorithm to run slowly,and can not achieve satisfactory matching results,so it is difficult to be applied to real-time tracking of targets.And although ORB algorithm runs faster,but because the descriptor does not have the scale invariance,so the matching effect is not ideal.In this paper,the feature matching algorithm based on sparse structure,the improved ORB feature matching algorithm and the target tracking method based on feature matching are proposed.The main contents are as follows:(1)On the basis of SIFT algorithm,a feature matching algorithm based on sparse structure is proposed,which aims at the time-consuming problem of traditional feature points detection.Inthis paper,the relationship between image blocks is found by solving the similarity of the image blocks,and the Sparsity function is defined according to this connection.The concept of sparse structure is proposed by the relation between the value of the Sparsity function and the image structure,and the structure area of the image is extracted.Only the feature detection of the structure area of the image is carried out to reduce the region of the feature detection and improve the speed of the algorithm.Finally,the error matching pairs in the matching results are eliminated with the RANSAC algorithm,and the accuracy of the matching is improved.The experiment shows that the improved algorithm has better advantages in matching speed and accuracy compared with the traditional algorithm.(2)An improved ORB feature matching algorithm is proposed to improve the ORB algorithm without scaling invariance.In feature detection,SIFT detection is used instead of FAST detection,so that the obtained feature points have scale invariance.The coordinate axis is established in the main direction of the feature point,and the BRIEF descriptor with rotation invariance is obtained.Finally,GMS algorithm is applied to compute the support matching of every matching point neighborhood,and the matching neighborhood is divided into the right and wrong matching neighborhoods,and only the matching pairs in the correct neighborhood are reserved.The experiment shows that the improved ORB feature matching algorithm has better robustness than the ORB algorithm.(3)The feature matching algorithm based on sparse structure is combined with the KCF target tracking algorithm.The size of the target is estimated by scale Pyramid,and the adaptive target size of the tracking frame is realized.During the tracking process,we judge the target occlusion.When the target is occluded,the template feature is extracted from the target in the current frame,matching with the features of the next frame,the target matching with the template feature is framed,and the target is tracked again.Experiments show that the feature matching algorithm based on sparse structure is applied to KCF target tracking algorithm to solve the problem of tracking loss due to occlusion and complex environment to a certain extent.