Research On Moving Object Detection Based On Sparse Background Modeling

Moving target recognition in video processing is an important part of many computer vision applications,and also used in many fields such as military,transportation and entertainment,with a very broad application prospects.In recent years,sparse representation theory has been widespread concerned by scholars,and some scholars have proposed robust principal component analysis models,dictionary learning models based on sparse representation.These two models have been functioned well in the field of video moving target detection,and achieved good results.In this paper,an in-depth study of the traditional dictionary learning models and traditional robust principal component analysis models has been given.Because the traditional models are easy to cause distortion and vulnerable to noises and along with many other issues,an improved algorithm based on K-SVD dictionary learning and an improved robust principal component analysis model combined with group sparse feature have been proposed to to alleviate the problem.The main work and research results are as follows:(1)Through extensive literature reading,a brief discussion about the research background and significance of this topic has been given,which introduces the present research status on moving target detection based on background modeling,sparse representation theory and video moving target detection.And systematically expounded the basic theory and modeling method of video moving target detection based on background modeling and foreground detection.(2)In-depth study of the traditional dictionary learning model,and puts forward an improved algorithm based on K-SVD dictionary learning of video background modeling method using discrete cosine basis dictionary as the initial dictionary,and by sparse coding get the sparse coefficients of the video in the dictionary,then by K-SVD algorithm update the dictionary in order to achieve video background updating,by background subtraction to detect the moving target.Simulation results show that the model in the background generation is affected less by the moving target than the traditional model.(3)In-depth study of the traditional model of robust principal component analysis,and analyzes the advantages and disadvantages of several commonly used solving algorithm of the model.An improved robust principal component analysis background modeling method has been proposed which based on traditional model and combined with sparse filtering group sparse feature,which firstly solves the robust principal component analysis model to obtain the background image of low rank and the sparse foreground image,and the foreground template which contains the moving targets is produced by taking the foreground image through a group sparse process.Simulation results show that this model is an effective solution to the effects of noise on the video moving target detection.(4)The summary of the work is carried out.And the areas in which video moving target detection still needs improvements are summed up.And looked to the future direction of the video moving target detection field.