Research And Application Of Video Processing Technology Based On Cloud Computing

Moving target tracking technology is one of the core technologies in video surveillance, its purpose is to extract velocity, acceleration, trajectory, the information of moving object in video image sequence to provide to a higher level of processing module behavior understanding and description, the stand or fall of movement target tracking results directly affect the high level of processing results.At present, ten of thousands of monitoring equipments produce a lot of surveillance videos every day, which gives us a great challenge in video handling works such as the browsing, storage and intelligent analysis.Cloud computing is an effective technology that can solve these problems,which can provide unlimited computing capacity and unlimited storage capacity.The technology not only achieve effective storage, but also be easier for users to intelligent analysis the surveillance video, which has become a hot technology of handing the surveillance video now.This paper firstly elaborates the core technology of the cloud computing.Then research the most popular Hadoop cloud platform now. The key technology of Hadoop is introduced, including the distributed file system HDFS, parallel programming model graphs, and the job scheduling mechanism in the Hadoop distributed system.And then the paper introduces in detail based on particle filter target tracking technology. The main work is as follows:For the massive surveillance video data storage and processing, based on the Hadoop system tracking a moving target is proposed in this paper. This system is divided into three layers architecture, which is IaaS layer, PaaS layer and SaaS layer.The Hadoop platform is located in the PaaS layer,and in the subsequent chapters in this paper focuses on the layer in the surveillance video data processing technology. The current the Hadoop system have not data structure to deal with the video data and have not video storage technology. Based on the Hadoop distributed file system frame of video data storage, video data parsing to the forms and graphs for the key value of the output data structure is proposed. Through the above design, make the Hadoop platform can directly monitor video data processing system, laid a solid foundation for further study on the following sections.The movement target tracking algorithm which under the traditional stand-alone mode can't parallel run on Hadoop cloud platform. The based on MapRduce parallel particle filter target tracking algorithm is proposed in this paper. The each of particle in each process is independent and have not data dependence,so this principle provide the possibility for MapRedunce parallelingthe the particle filtering target tracking algorithm. The experimental results show that the parallel particle filter target tracking algorithm based on Hadoop improve the calculation efficiency when compared with the existing algorithms improve the calculation efficiency.In order to improve the execution efficiency of processing video data in the Hadoop cloud platform furtherly, Hadoop task optimization strategy is proposed in this paper,which is the length of the adaptive adjustment of two stage queue scheduling algorithm. In the algorithm according to the different types of video monitoring data(CPU intensive or memory intensive) match the corresponding type to execute Hadoop cluster corresponding node, and the length of queue based on the system operation of each node in the process of running state and adaptive adjustment,ensure the cluster can be the most efficient task scheduling and task execution.Experimental results show that compared with current existing Hadoop scheduling algorithm, this paper proposes an improved scheduling algorithm to improve the execution efficiency of platform.