Site Loader

1. This is a framework to show how to process, organize, manage and store massive video data. The framework has three parts: video intelligent analysis (Object detection, target tracking, behaviour analysis and event analysis) and video structured description (VSD) are utilized to mine valuable information (persons, cars, unusual behaviours etc.) from large scale video data, which then is expressed in standard format. 2. The second part is construction of policing repository database, which is used to data mining, information describing, moreover knowledge reasoning as the domain knowledge, and provide real cases to assist crime prediction. Furthermore virtualization and cloud computing provide efficient computing environment for techniques all above, and storage environment for various types of structured and unstructured data.3. For types of tasks such as video content analysis, semantic modeling and reasoning, MapReduce, Spark, Storm and other distributed processing model are applied to deal with corresponding task. Take video retrieval for example, MapReduce would be used to support the task, of which the key is represented by the time in video, and video data are divided into several parts by the key, then all tasks execute simultaneously.All the data are packaged with unified standard format and transferred to the distributed cloud platform which provides greatly efficient storing and computing ability. Due to the limited bandwidth, the “front + back” pattern is adopted, that is: simple video analysis algorithms are carried out in the cameras, and results are sent back to “the cloud” to support more complex computing and applications. The pattern could avoid network congestion caused by large-scale video big data.Repository database could be constructed as follow steps:1. knowledge collection, that is collecting and analysing existed cases, policies and regulations, and make them as knowledge repository sample set; 2. knowledge discovery, that domain knowledge are mined, clustered and analysed from the collected cases and rules, with machine learning such as support vector machines (SVM), or expert guidance;3.  Knowledge representation, domain knowledge and rules should be represented with unified form such as RDFS, OWL and SWRL, and stored in repository database and model database, from which the information would be utilized to support training models, semantic retrieval, reasoning and crime prediction. 

Post Author: admin

x

Hi!
I'm Sonya!

Would you like to get a custom essay? How about receiving a customized one?

Check it out