The Research On The Algorithm Of Moving Target Detection In Complex Background

Moving target detection is a significant part of the research of computer vision,it is the basic of target tracking,target classification,target behavior understanding and many other steps of video analysis.The technology of moving target detection can be used in intelligent manufacturing,security system and automatic production.In the actual scene,the video usually contains illumination variations,motion changes,shaky leaves,low contrast et al,namely dynamic scene.How to extract the moving target in the dynamic video accurately is the main problem we need to deal with in our study.In order to adapt to the change of the scene better and increase the accuracy rate of target detecting,the paper improved a method of self-adaption threshold and Ghost controlling background subtraction.Selecting the normalized RGB value as the characteristics of the pixels.For saving time in the process of initialization,the first N frames were selected as the initial background models.During the process of background subtraction,we focused on the change of contrast and adjust the threshold of the subtraction.Moreover,random updating scheme was employed here to find a balance between the stability of the background models and its adaptation to the variation of the scene.Finally,the parameter of foreground lifetime was introduced to solve the Ghost problem,those pixels whose lifetime was longer than the threshold would be classified into background.The experiment of target detection was carried out on static scene,Wallflower data set and SABS date set.The result showed that the proposed algorithm could gain preferable accuracy rate.On three scenes of the Wallflower data set and SABS set,the F-score of precision and recall got best results,respectively were 69.08%,80.69%,75.43% and 74.51%.What's more,it was very efficient,which meant it was a real-time method,the detection speed is 11.21 fps,20.35 fps,9.96 fps respectively.Furthermore,we also searched the influence of the post-processing algorithm on the result of foreground extracting,the result indicated that suitable post-processing method could improve the result of target detection obviously.Then,based on the realization of multiple algorithm,we exploited a moving target detection GUI using MATLAB.The function of the GUI includes file opening,algorithm selecting,real-time detecting result displaying,time calculating and so on.The use of the GUI can improve the automaticity of target detection,the function zone of it is clear and the background program is easy to modify.It could provide qualitative and quantitative result for algorithm comparison.Finally,we selected a specific scene–a heavy foggy day to do the target detection experiment to investigate the practicability of the algorithm.First,we analyzed the characteristic of the incoming image and the reflection image and then improving a defogging algorithm based on vision theory.After video defogging,the information entropy changed from 7.41 to 7.61,improved by 2.7%,the contrast ratio was increased.Then we extracted the reflection component to realize video defogging.After that,detecting moving object.The result showed that the proposed method is practical in real scene.