Study On Target Detection Algorithm Based On Weighted SIFT Feature

Target detection in complex images is a hot topic in the field of vision,which plays an important role in the fields of industry,military and intelligence.With the complexity of the current image,target detection due to background scene complexity,target attitude change,light is not balanced,the target and background are blocked and other reasons,has become a challenging problem in machine vision.Although the target detection technology has been studied by many scholars at home and abroad,but there is not a general and robust target detection algorithm.In this paper,we focus on the target detection methods in complex scenes.Specific research contents and results are as follows:(1)A target detection algorithm based on weighted SIFT feature matching is proposed based on the characteristics of the three main target detection algorithm,weighted SIFT features: first,in the training phase is proposed using the weighted template to describe the target information in the template image distribution,to distinguish the importance of different target template image information using weights,can filter most of the background information and interference information extraction information of the target through the middle of visual elements of mean shift clustering can be distinguished here,not only can express positive characteristics of the sample,and has a very good distinction for the negative samples.Secondly,we use the SIFT feature to describe the template image and the detection image in the detection stage.Finally,this paper puts forward a method to generate the matching frame based on the matching point mapping.By comparing the weighted SIFT features with the original SIFT features,this paper shows that the weight template improves the SIFT feature matching.The effectiveness of the algorithm is verified on the target detection algorithm,weighted SIFT feature and the target detection algorithm,experiments show that this algorithm compared with Haar-adaboost algorithm and HOG-SVM algorithm has higher precision and accuracy.(2)This paper proposes a method of detecting frame based on matching point density detection of matching point density frame generation algorithm based on the improvement of the frame generation module detection target detection algorithm,weighted SIFT feature based on the use of the module matching point matching frame generation method based on mapping,mapping,because frame generation algorithm is more sensitive to the changes of target attitude based on the matching process,so as to cause error detection.The feature matching density detection point for SIFT frame is the essence of image feature matching points of maximum density area based on this method,based on statistics,considering all of the matched points,thus for some point mismatch or because a target attitude change matching point displacement have a very strong stability.By comparing the experimental method and frame generation method based on mapping,proved the validity of the method.In this paper,the general target detection technology under complex environment is studied systematically,and the detection algorithm based on weighted SIFT and the detection frame generation method based on matching point density are proposed.In this paper,we give some solutions to the problems of complex images,and get good results.