RIMS: Real-time and intelligent monitoring system for live broadcast platform

Personal live shows on Internet streaming platform currently is blooming as a popular application and attracting several millions of users. The content supervision of live streaming platform, in which hundreds of show rooms perform synchronously, is a major concern with the development of this new kind of service. Traditional image captures and realtime content analysis experience huge difficulties, such as processing delay, data overwhelming and analysis overhead. In this paper, we propose a new method to monitor real-time live stream and identify illegal live misbehavior intelligently based on state of perception and frame difference analysis. The proposed system makes use of several indicators of the chatting room status rather than analyzing images solely and directly. Three kinds of detecting techniques are adopted: self-adaptive threshold-based abnormal traffic detection, sensitive comments perception, and state evaluation. The proposed system captures the segmentation of video scenes by frame difference analysis, and pays more attention on the showing room that has large scene changes. We deploy our system to monitor a typical live platform called panda.tv, and the overall accuracy of three indicators reaches 90.1%. The application of this system can change current situation where the live platform supervision totally depends on manual review. The key techniques in this system can be widely ∗Wei Ren,weirencs@cug.edu.cn Preprint submitted to Elsevier April 10, 2018 employed in many applications, such as live broadcast platform industries, video surveillance for state security, and the national security for counterterrorism.