Research Of Image Matching Based On Region Block And Applications In Complex Scenes

Image matching is widely used in target detection,security monitoring and motion estimation.The same target has great difference under the influence of illumination change and so on,so the image matching and application in complex scene is a challenging problem.The target recognition method based on region block is a typical method in the field of motion tracking.This paper applies to static image matching and tries to solve the problem of image matching in some complex scenes.Considering the influence factors such as illumination change and noise in complex scene,this paper,starting with the structure information and locality of the block,combined with the Best-Buddies Similarity matching(BBS)algorithm,gives a better robust algorithm for illumination change,noise,scale change,rotation and angle change.In addition,because the human body can be seen as a series of regional blocks(namely human limbs,head),but these regional blocks have different shapes and relative positions due to the influence of angle and illumination in complex scenes.We use the hinges model of human posture estimation to match these blocks through machine learning methods,and use the relative position to identify the human target.The research in this paper mainly includes:First,combining BBS algorithm and block matching idea,a matching algorithm based on region block preprocessing is proposed.The algorithm is preprocessed by the filter of the region of interest,which makes the image to be matched under the influence of illumination change and noise,and still has good matching precision.In terms of matching criteria,instead of extracting the feature points of the region blocks,the distance function is introduced to match the pixels directly.Because only local areas are considered in this way,simulation experiments show that in complex scenes,image points based on region block preprocessing have higher matching accuracy and robustness than previous matching algorithms.Second,based on the idea of articulated component model proposed by Yi Yang et al,a multi-target human detection algorithm in complex scenes is proposed.The algorithm uses the traditional HOG algorithm to extract the feature of the region block.On this basis,the Support Vector Machine(SVM)is used to train and identify the parts and estimate the relative position,and then the Gauss model segmentation algorithm is used to identify the human target in the complex scene.The simulation results show that the multi-target human detection algorithm in complex scenes is better than the previous recognition accuracy.