The Research And Implementation About Video Data Mining With Parallel Computing Model

With the rapid development of Internet,video services of social media and video sharing of social networks popping up have made the number of network video increasing rapidly.Storage,query and analysis for massive videos is a major challenge in the field of computer science and technology.Cloud computing,as current popular technology,has a natural advantage in massive video data processing.As an emerging parallel computing framework based on memory,Spark has great potential and broad application prospects.On this background,this paper made the following three work in the research of parallel video data mining.This paper proposed a large scale video data mining architecture,which is based on Spark and other open source software and has three layers.Parallel algorithms about feature model of video data were researched in this paper,including shot boundary detection,key frame selection,feature extraction,visual vocabulary generation and bag of features etc.On the basis of parallel feature model,we proposed two video data mining applications,human action recognition and near duplicate video retrieval.To address human action recognition problem,we constructed a classification model based on spatial and temporal features of video,using SVM classifier to solve it.And to address near duplicate video retrieval problem,we designed and implemented an efficient parallel retrieval algorithm based on key frame local features.This paper conducted relevant experiments to verify the parallel video data mining framework's performance.Respectively,we conducted human action recognition and near duplicate video retrieval experiments on multiple benchmark datasets.Using Average Accuracy,MAP and other indicators showed the effectiveness of video mining framework,and using execution time and speedup demonstrated the high performance of framework.Finally,this paper introduced the function and architecture of big data analysis platform(BDAP),which is developed by our laboratory,and described the design detail of parallel video data mining algorithm in an integrated into system process.At last,using human action recognition and near duplicate video retrieval application cases verified the availability of integrated components.