Design And Implementation Of Parallel Processing Platform Of Video Base On Saprk Streaming

The surveillance cameras are distributed in every corner of the city,the number of surveillance video has been explosive growth.The traditional online intelligent monitoring video processing system assigned video stream acquired by one or more cameras to an intelligent video unit,and then the intelligent video unit make video decoding and video analysis on the streaming.Because the video analysis service is highly couple to the video processing resources,the system can only takes advantage of a few resources to deal with a lot of requests of the same services,which reduces the utilization rate of resources.Video reading tasks is highly coupled to video processing tasks,the intelligent video unit failure will cause the video data loss,data reliability is not high.According to the problems existing in the traditional on-line monitoring video processing system,this paper studies study effective parallel processing platform for surveillance video.At first,this paper analyzes the requirements and designs the architecture based on the Spark Streaming.Adding a video streaming aggregation module based on the Kafka in front of the video processing module.This module decoupled data reading and processing,and caching video streams,for improving the reliability of data.In the video processing module,the paper uses Spark Streaming distributed platforms technique,ransparentizing the computing nodes and making algorithm processing parallel to reduce resource fragments.Then,according to the whole design of the monitoring video online parallel processing platform,this paper develops the whole platform.In the video processing module based on the SparkStreaming,the paper proposes and implements KafkaVideoUtil utility class,this class make the process of reading and parsing Kafka video messages process transparent,and Simplifies the complex distributed video data acquisition and analysis work to improve the efficiency of the video analysis application development.Based on this platform,this paper designs and implements the traffic surveillance video vehicle metadata extraction system.Finally,the paper test the platform,including the reliability testing of the core modules and contrast testing with the traditional on-line monitoring video processing system.Experimental results show that the platform has high reliability,the number of system resources pieces are reduced,the resource utilization is improved.