Design And Optimization A Video Processing Framework Based On Distributed Computing Model

With the rapid development of Smart City,Video Technology had been widely used as a basic data collection method which had caused a problem: With the passage of time,more and more video data would produced per unit time.The traditional methods of processing massive video data hadn't satisfied the demand of new situation.So,the massive video data which generated in a short time wouldn't process promptly was serious affect the timeliness of data value.In this paper,we proposed a video processing framework based on distributed computing model.It supported multi-client updated videos simultaneously,then extracted faces which appeared in those videos and built an index table which store in memory to increase query speed.Through processing a frame image which uploaded from client had special faces,the framework could locate those faces in those videos.The content and innovation of this paper are as flower:Firstly,this paper divided the overall framework as function modules and describing the service algorithms which may be involved during the framework developing and choosing the Best fit algorithm form what we had.Meanwhile,researching the involved algorithm which is process faces.Analysis the advantage and disadvantages of various implements and picking a fit one.Secondly,Basing on function modules and software which is selected to implement the framework,we described each design details and optimized it from the perspective of entire cluster: 1.How and why the data stored.2.Design distributed tasks based on MapReduce computing framework.3.Analysis the performance degradation factors during the cluster running and finding appropriate solutions to prevent or avoid the occurrence.Then,the whole cluster can run stable under heavy-load situation.Finally,repeatedly an experiment that retrieve one special face in two million images data were used to reflect the effect of this framework.Eventually,the average retrieve time by using this framework to find a particular face is 665 ms.According to the experiment,this Framework has good performance and actual demand was satisfied completely.