| Video data is rapidly expanding with the popularity of video surveillance, making it difficult to monitor for 24-hour totally by labor. Intelligent video surveillance system can achieve real-time object detecting and tracking under unattended operation, analyze the behavior of target for real-time event detection, and has become a hot research direction in the field of security surveillance. However, the existing intelligent video surveillance system is faced with real-time problem because of the complexity of the intelligent algorithm and the popularity of high definition video.As traditional intelligent video surveillance system has poor real-time, this thesis designed implemented software for Davinci intelligent video terminal, which realizes image acquisition, video denoise, video analysis and other functions. Video processing is divided into modules on each subprocessor according to the characteristics of Da Vinci, and a communication mechanism between the multi-core tasks based on Mailbox is developed for message exchange between modules. Down-sampling BMA is adopted to improve 3D video denoise algorithm, which reduces the processing time while keep image quality. Video analysis framework utilities multi-thread design to separate the business layer and algorithm layer, which ensures the system’s real-time. Moving human body detection under fixed scenario is implemented using background difference method for this framework. To avoid performance bottleneck problems in traditional server when manage multiple intelligent devices, the monitoring information processing platform uses Storm flow processing engine is developed, and reduce the message processing delay. Experimental results show that the system can meet the real-time need for high definition intelligent video surveillance, it can achieve 30fps video output with a 85ms terminal processing delay, and can achieve real-time processing for mass monitoring information. |
PDF Full Text Request