| How to locate the effective data quickly in the massive surveillance video data has become the urgent demand of the security surveillance profession.Video synopsis is an effective way to meet this demand.It extracts the effective information from the original video to reconstruct and realizes the purpose of browsing important information quickly.This paper studies the key technologies of video synopsis,including background segmentation,multi object tracking and object combination optimization algorithm.In order to improve the accuracy of video synopsis,effectively reduce the information loss rate,improve the concentration degree,this paper improves and optimizes the key technology,the main work includes: In the background segmentation,considering the robustness of the LBSP characteristics of the light,the background modeling method of non-parametric LBSP estimation based on feature was put forward.At the same time,in order to cope with the challenge of foreground extraction in different scenes,the traditional Gauss mixture model is improved and fuses with former method.In the aspect of multi object tracking,the multi object matching tracking algorithm is used to model and feedback the scene,and the collision and occlusion objects are marked to preserve the real collision in the original video.In the combination of object,video synopsis is classified as single synopsis,single segment synopsis,multi object synopsis and multi object segment synopsis according to the number of objects and segment for the first time.Objects are classified by unsupervised training and supervision training based on SIFT and motion features,and finally realize the specific video synopsis,then video retrieval is implemented after that. |
PDF Full Text Request