Detection And Tracking Of Weak Targets In Complex Background Based On Improved Compressive Sensing Algorithm

Target detection and tracking is regarded as a important research topic of computer vision,which combines image processing,pattern recognition,artificial intelligence and many other disciplines.Especially for the small and weak target detection and tracking under complicated background has already become a current research hot spot.The key and difficulty in this study is how to improve the accuracy and efficiency of target detection and tracking.This paper focuses on the compressive sensing algorithm,and do some research from three aspects:image pre-processing,target detection and target tracking.The main research contents of this paper as follows:(1)A large quantity research has been done for the feature selection problem in the target tracking algorithm based on compressive sensing(CT algorithm),especially for the native bayes classifier model H(v) in CT algorithm.It can be obtained that each feature corresponds to the h(v_i) has parabolic characteristics after formula derivation and theoretical analysis.h(v_i) is a function of the mean and standard deviation of the positive and negative samples,the maximum of h(v_i) can be obtained at the axis of symmetry when the standard deviation of the negative sample is greater than the standard deviation of the positive sample.In this case,the difference between positive and negative samples is high.So this feature has better classification effect.Based on this,a feature selection method which according to the negative sample standard deviation is greater than the sample standard deviation has been proposed.The validity and accuracy of the improved algorithm has been proved through a lot of numerical experiments with Open CV function library and Visual Studio 2010 as experimental platform.(2)For the weak target tracking,acompressive sensing target tracking algorithm based on feature weighting(WCT algorithm)has been proposed in this paper.Firstly,the characteristics of weak targets and the difficulty of research are analyzed.Then,the mathematical theory of histogram equalization is analyzed from two aspects:enhancement function continuous and enhancement function discretization.The image has been processed by histogram equalization,so that the pixels distance between the object and the background in the image is widened.That is,the difference between the mean of the positive sample and the mean of the negative sample becomes large,and make the image contrast enhanced.The characteristics of the image are more discriminative.Last,the feature weight is calculated by the method of the feature optimization algorithm of Relief(Relevant Features),and each feature is assigned a corresponding weight.The weight is updated after each frame gets to the target.The WCT algorithm is compared with the CT algorithm and the GCT algorithm.It is proved that the WCT algorithm can track the weak targets accurately and in real time,and has good robustness.On the one hand,this paper gives a feature selection scheme based on na?ve Bayes classifier model H(v)and improve the tracking effect of the compressive sensing algorithm;On the other hand,based on the idea of image enhancement and feature weighting,we propose a new small target tracking algorithm and achieve an ideal tracking effect.