Video Condensation And Classification Based On Object

Video condensation is a new technology that aims to deal with massive video data in modern society.With the video being analyzed,this technology is able to remove the redundant part of the video and cut back the length of the video.In order to expand its application,we propose a method named video condensation and classification based on object for finding the key information of objects in surveillance videos and condensing videos according to user's option.Following is the main tasks of this paper:1.In this paper,we propose a method of video condensation based on foreground clips which allows user to specify objects that have particular characteristics in condensed videos.Foregrounds of diverse objects in different space and time are merged into the same scene on the basis of above method so that the length of the surveillance video can be reduced greatly.2.Analyze the background modeling methods and features,then choose the best method and features to separate the foreground and the background from the surveillance video.3.We propose three foreground pre-processing methods to purify the foregrounds yielded by background modeling.The method of shadow suppression based on SILTP achieves good results since it considers both color and texture features in shadow region detection.4.Four kind foreground information are proposed,including position,color,direction and type information.Meanwhile,a high speed object tracking method is proposed to unite the information of objects so that the influence of "blink" effect among objects can be reduced in condensed videos.5.A fully functional system is built according to this paper,which is able to condense surveillance videos in a particular condense rate or classification.System performance has basically reached the relevant standards.