Video Analysis For Task Management And Execution System

With the continuous development of the national economy, urbanization, more and more attention to the construction of urban and campus video surveillance and analysis. But the video analysis platform met two important technical problems: First, a large number of parallel processing capabilities of video analysis tasks; the second is the different types of video analysis tasks scalability.For these two technical problems, this paper proposes a system to support a variety of different types of video analysis task management and execution, which based on an actual project. The task management is extracted as a Control Server, responsible for video analysis tasks load balancing. The machine which performs video analysis tasks is called Job Server, mult IPle Job Server work together to make full use of CPU computing power and solve the problem of CPU-intensive computing. In addition, Control Server is the communication bridges between the Web platform(responsible for registration / cancellation of video analysis tasks, display alarm events, etc.) and Job Server, responsible for collecting load information from the Job Server periodically and forwarding and processing registration message to the load lighter Job Server.To ensure efficient communication among Control Server, Job Server and front-end Web platform, we design a communication protocol based on TCP byte stream. Video analysis tasks for Control Server is concerned as a packaged TCP packets based on communication protocols, Control Server is only responsible for the distribution of video analysis tasks to the target machine. As long as each different video analysis algorithms bases on the rules of the docking module, providing a running entrance for the task execution, the incoming interface of picture frames and digital outcome feedback, it can easily perform hot swappable in the system.In order to ensure the robustness of the system, the Control Server uses the plan of hot standby disaster recovery. When the host goes down, it can quickly switch to the backup machine and continues to work. In the Job Server layer, when a machine goes down, the Control Server will run a migration to the other machine, in order to ensure the system can provide service as usual. In addition, the paper also uses a scr IPt to monitor the service process, in oeder to safeguard the stability of the system and restart the process when an exception occurs.Finally, this paper will do some experiment on the key modules which influcence the system performance, and analyzes the experimental results to verify the efficiency of task load balancing algorithm and effectiveness of the communication protocol.