| As the country’s requirement for urban safety construction gradually improved,intelligent security technology has become an important means to guarantee urban safety.Intelligent monitoring products of real-time target detection and event mining realized by computer vision technology have been gradually deployed to each monitoring and security network of the city.However,with monitoring data generating in quantity,how to store the massive video data and how to analyze the effective information rapidly from the massive video data has become another problem now in the construction of safe city,therefore this paper opens up discussion on the basis of these problems,processes distributed video data by use of Hadoop,the mainstream framework in the field of big data is the main frame in data field,sets up face recognition system based on Hadoop,and proves by experiments that the system could greatly improve the processing capicity of video data.In this paper,by combining Hadoop technology and computer vision technology,face data mining from the massive videos is carried out,and the mined data is managed for the fact retrieval later.This paper first analyzes in detail the structure and principle of big data technology,the Hadoop framework,distributed file system HDFS,distributed computing model MapReduce and distributed database Hbase.Then on the basis of the above analysis,the paper proposes extending Hadoop data type interface,Hadoop file input and output format interface and Hadoop file fragmentation interface to achieve Hadoop’s support on video data processing.In order to solve the problem of video frame losing in the upload process of video data to the distributed file system HDFS,video data is segmented according to the size of HDFS file chunk in the upload process by use of FFMPEG multimedia framework,and then uploaded to the distributed file system.In the map function of parallel computing model MapReduce,face detection algorithm based on Adaboost and face feature extraction algorithm based on PCA are used to recognize face information,and the recognized information is summarized and uploaded to Hbase database in reduce function.On this basis,this paper retrieves the face information data stored in the Hbase database.Finally,this paper employs the video data to test the function and performance of the whole system.Experiments show that the Hadoop-based face recognition system can more than double the performance,compared with the standalone processing method.Due to theextensibility of Hadoop,the system performance still has a lot of room for improvement by expanding Hadoop cluster. |
