The Research And Application Of Anomaly Detection Algorithm For Visual Big Data

Under the modern society,with the rapid development of science and technology and economy,the Internet and multimedia technologies have been rapidly iteratively updated,and the era of visual big data has really come.This has a major impact on the field of intelligent surveillance and presents a huge challenge to the field of intelligent monitoring.In the field of intelligent surveillance,people hope that monitoring can truly achieve unsupervised,fast response,and intelligence.The traditional surveillance video system is faced with anomalous algorithm analysis that can not perform real-time analysis on high-resolution surveillance cameras.Usually,a large amount of manpower and material resources are used to ensure the security protection in the surveillance field.This not only wastes human resources but also is affected by human factors.Therefore,how to quickly and efficiently obtain the information required by the user from high-definition real-time video data has become a research hotspot in the field of intelligent monitoring.The traditional stand-alone architecture has not been able to meet the current real-time requirements.The current distributed computing has become the mainstream computing model of big data.Therefore,this paper integrates distributed computing with the background modeling method,focuses on anomaly detection algorithms for large-scale visual data,and makes full use of the Spark distributed framework to improve the efficiency of the background modeling architecture and completes the following aspects:work:(1)In this paper,several traditional anomaly detection algorithms are deeply analyzed.The background difference method,inter-frame difference method,optical flow method and background modeling method detection principle and process are summarized and compared,and the advantages and disadvantages of each method are described.Secondly,the robust principal component analysis methods are reviewed,including the description of static robust principal component analysis theory and the comparison of several "online" methods of dynamic principal component analysis.(2)Two background modeling methods based on RPCA are proposed in this paper.They are average grid background modeling method and multi-ROI background modeling method.Combining with the defects or bottlenecks in the traditional anomaly detection,the principle proved the feasibility of the two methods,and the application of the two methods in the anomaly detection process was elaborated in detail.Finally,the experimental results show that under the premise of ensuring the accuracy of the two methods,the detection time is greatly reduced.(3)With the advent of ultra-high definition camera monitoring,the challenges faced by intelligent surveillance are extremely prominent.In order to meet the requirements of real-time detection,this paper proposes a distributed implementation of Spark clusters for the two background modeling methods above.Theoretically proved the feasibility of parallel processing based on Spark,and detailed deduction of the parallel algorithm flow.Secondly,the high-definition surveillance video is based on the implementation process of parallelization of Spark clusters and the feasibility of the parallelization process is demonstrated through experiments.Finally,the detailed configuration of the Spark cluster and the experimental results prove that compared to the two background modeling methods mentioned above,the background modeling of ultra-high-definition online surveillance video based on Spark clusters The detection time of the law is reduced more and can meet the real-time requirements.