Font Size: a A A

Research On The Video Big Data Of A Intelligent Early Warning Application Based On Hadoop

Posted on:2016-12-21Degree:MasterType:Thesis
Country:ChinaCandidate:P XiFull Text:PDF
GTID:2308330479998354Subject:Computer technology
Abstract/Summary:PDF Full Text Request
With the growing awareness of social security, video surveillance, intelligence analysis and real-time early warning, as means of security systems, are widely used in every corner of society. But the uninterrupted operation of the monitoring system and the unstructured video data will produce massive and complex video data, which makes the traditional tools to manage and analyze the big data inefficient. Hadoop is a popular open source about distributed computing framework. In this paper, I tried to combine the intelligent early warning system with the framework to achieve the video retrieval of parallel processing and distributed file storage, which can solve the problems about low efficiency of handling, large storage and complex structure.The main research of this paper is to use the Hadoop for parallel computation and distributed data stored for content retrieval of large amount of video data, to retrieve similar videos by uploading image in the service of the intelligent early warning system. First of all, use the key frame extraction mode to convert each video into a plurality of video frames, then obtain the numerical feature of each frame by using the SIFT feature extraction algorithm On the basis of that, we can match the SIFT feature with the retrieving image to obtain retrieving results. Combining the independent running video frame SIFT feature extraction with MapReduce parallel model, defining the input/output values and Map/Reduce task content, can calculate the multiple frame feature values simultaneously. In addition, we can reduce the calculation range to avoid the matching calculation between retrieving image and each video frame SIFT feature value, by LSH local sensitive hash map, divides the SIFT feature vectors with high similarity into a group, when retrieving a image only searches the videos in the corresponding group, Stores data generated during the analysis by HBase, designs the data list structure for retrieving video easily, which can realize the efficient video retrieval function. Finally, this paper verifies the promotion of the retrieval performance by experiments.Using the Hadoop architecture to achieve the video retrieval function of the intelligent warning system, not only can improve the retrieval performance but also has the advantages of high reliability, good fault-tolerance, which provides the foundation of the parallel design and implementation of other functions in the system. It is an important way to solve the security problems of big data.
Keywords/Search Tags:Hadoop, video retrieval, SIFT feature extraction and matching, LSH, HBase
PDF Full Text Request
Related items