Complex Dynamical Background Extraction Method Based On Time-varying Filter

Extracting background from video sequence is a kind of basic technology.It is not only applied to moving object detection,but also widely used to background modeling,object tracking,foreground detecting,privacy camera,intelligent technology and so on.Applications involve video surveillance,traffic flow detection,intelligent robots,unmanned aerial vehicle intelligence acquisition,camera technology and so on.Human vision have limitations which includes unclear targets,too many targets,fast target movement in complex scenes.Those will change the results of observations and the judgments.It consumes energy and time of people because of wrong judgments.And it does not match the real-time by artificial monitoring.So accuracy of the monitoring may not be proved.At present,there are many problems in background extraction which include the inaccurate detection of the target,the influence of weather illumination in the process of background modeling and the background extraction in complex scenes.Due to the above factors,the result of the background extraction can not avoid blur,ghosting and noise.Because of background images extracted incompletely,we can not achieve the target motion tracking,nor can it be applied to the background modeling and privacy camera.This paper will analyzes and solves the above problems in the background extraction processes and results.It will propose a background extraction method of time-varying filter to obtain a more complete and accurate background model.And it can improve the robustness of the background extraction results and reduce the background model Signal-to-noise ratio.So makes the background model extraction process more effective.Based on time-varying filter background extraction is used in a period of dynamic video.The time-varying filter can detect the foreground of the movement,and remove the foreground of background extractions.The time-varying filters are used in background modeling and background updating.The purpose of this method is to detect the foreground of motion and extract the background in a period of video.The method we proposed in this paper consists of two steps: the first step is to establish the original video sequence image samples.It is estimated by double-threshold improved kernel density algorithm,Then weight samples after sample composing.The second step is to establish the background space,andcollected samples are projected to the background by space.After Vector space background extracting,background model reduce dimensions.And we can get the final image background model.The algorithm can be up to 99% precision in land dynamic background and up to 85%in background which contains water.The extracted results can effectively avoid the failure of dynamic background extraction and illumination.