Content-based Video Summarization And Retrieval System

In recent years, with the strengthening of the security field in our country, monitoring technology is being greatly developed and the number of surveillance cameras in Chinese cities continues to increase, covering a lot of areas such as streets, public places, office buildings, uptown and so on. On one hand, the increase of the number of surveillance cameras makes a contribution to reduce crime and plays a key role in assisting the police to break cases by providing supporting evidences. On the other hand, the increase of cameras have also brought more stress to monitoring staffs who will have to pay more energy and time to deal with massive amount of real-time video data. The monitoring staffs have to watch a 24-hour video if they want to check out the abnormal events within a day. However, most of the time, there won’t be any people or cars occurred in video scene such as aisle, hall and country road. In most cases, the monitoring staffs will try to fast-forward to reduce playback time. This results in a waste of human resources and the strong possibility of missing important information. The problem is becoming more and more serious with the rapid expanding of the number of cameras. Therefore, it is an effective and necessary way to watch a high-enriched video instead of the long original video. Moreover, using image retrieval technology to discover suspects is also an effective auxiliary method. Above all, a video summarization and retrieval system will have great market prospect.Considering the characteristics of the monitoring scene, this paper designs and implements a content-based video summarization and retrieval system, which will generate a video summary for users through analyzing real-time surveillance video. The users can not only spend much less time to know all the information through the video summary, but also set some rules to filter target objects they interested in. And the system can also help the users to locate suspects through image retrieval. In this paper, considering the feature of the scene of surveillance, we present a method to generate a video summary for a real-time surveillance video by rearranging moving trajectories through time-space compression after extracting moving objects and calculating their trajectories. The moving objects will become denser in the video summary so it will take the users a shorter time to watch it. Algorithms such as background modeling, multi-target tracking and cross-border detection are used in this paper. And image retrieval based on color histogram is also used to retrieve target objects in surveillance video.The content-based video summarization and retrieval system has been used by a company in Guangzhou for several months. The experiment results show that the system can generate a summary by extracting the useful information in the video effectively and also can retrieve target objects exactly. It is proved that the system has good real-time and robustness characters.