| With the rapid development of economy,social security has attracted more and more attention over time.Improving the level of video surveillance and promoting the application of emerging technologies in the field of video surveillance have become and will continue to be the focus of social security work for a long time.Video Intelligent Processing aims to solve the current situation of human supervision and analysis of video surveillance.It integrates computer vision and image analysis technology into video processing system to meet diverse needs.Intelligent multichannel video processing has the traits of large amount of data and high complexity.Facing the real-time requirements and the limitation of hardware and software resources,it is of great significance to propose a solution of intelligent multi-scene video processing system.The main work of this paper is as follows:According to the repeatability and time sequence of video processing,the MSSM video scheduling management algorithm based on concurrent pipeline is put forward.In order to ensure the expansibility to different application scenarios,the algorithm is deployed in the pre-processing process of general video front end,including the overall parallelization scheme design,the algorithm architecture of the concurrent pipeline,the video scheduling algorithm,video storage management and video intelligent frame skipping algorithm,etc.,which is called video scheduling management algorithm(MSSM)based on scenario scheduling.Combining with the prior knowledge of application scenarios and the real-time feedback of video intelligent processing,the tasks and data are scheduled and managed,so as to achieve the goal of reasonably scheduling system resources and maximizing information of video processing.Based on the introduction of scenario factors of scheduling management algorithm,this paper studies and optimizes the video intelligent processing algorithm based on lightweight network for the application scenario of single-frame population statistics: Mobilenet is used as feature extraction network,small-scale features and global information are introduced,optimized.SSD is optimized as back-end detection framework to balance accuracy and efficiency.The algorithm can meet the requirement of instantaneity as the data foundation of the scheduling management algorithm as well as provide the reference by the deep learning model to deploy in the edge terminal,achieving lower latency,fewer model parameters and better robustness.Finally,aiming at the multi-scenario video intelligent processing and algorithm design related to scheduling management proposed in this paper,a solution of multiscenario video intelligent processing system based on edge computing is proposed.By building multi-channel video intelligent processing system based on single-frame population,extracting the algorithm module or scheme design that needs to be verified,combining with the corresponding software and hardware optimization technology,testing the system performance and effect,and analyzing and summarizing the results,the goal of high performance and stable operation of multi-channel video processing system is achieved. |
