Research On Synopsis Of Massive Video Data

With the rapid development of Safe City and Intelligent Transportation, video surveillance ispopular in areas such as transportation, schools, homes and shopping malls. How to quickly browselong videos,condense massive video and retrieve needed video data have become an extremelyimportant research field.Key-frame based video abstraction and object-based video synopsis are studied in this theses,the main work and contributions are as follows:1. The theses illustrates the research background and research status. A wide range of surveyof current dominant video abstraction approaches has been conducted. The technical principles，difference and status of research of key-Frame based video abstraction and Object-based videosynopsis are introduced.2. According to the shortcoming of K-means which need to set the number of clusteringmanually, an improved approach for K-means with auto selection of K value according to thecharacters of video content is put forward. By experiment the effectiveness of the improved K-means algorithm has been verified.3.An approach combining dynamic model detection algorithm Codebook and interactive objectsegmentation algorithm Grabcut is presented.Automatic moving object detection and segmentationis realized without human interaction.4. Aiming at the situation that the objects in security monitoring video are mainly pedestrian,car and cycle, a video object feature selection method is proposed. By LibSVM, the effectiveness ofselected object features have been verified, with the classification accuracy reaching80%.5. On Ubuntu platform, the development environment is set up with QT+OpenCV.The overallframework of software system, the principles and code implementation of core algorithms areintroduced in detail. After debugging, summarization and retrieval of security video has beenimplemented. The length of the synopsis video can be compressed to about10%of the originalvideo.