Research On The Construction Of Knowledge Graph In Video Detection

With the construction of the safe city,the public video surveillance system has played an important role in social security prevention and control,criminal crimes,intelligent traffic command,risk prediction and early warning.Its huge and rich video image resources provide clues for public security case investigation,assisting the investigation of police to quickly locate suspects,set case evidence,restore case detection sites,improve the accuracy and efficiency of case detection.With the rapid development and application of video surveillance systems,in the face of massive video image resources,how to find the involved information efficiently,accurately and completely is an urgent problem to be solved in the current video detection work.With strong semantic relevance,Knowledge Graph is a structured knowledge description,which can better organize and manage data.In the video detection,Knowledge Graph can further improve the efficiency of investigation,fully exploit the clues,and transform the resource advantages of massive video information into real-time combat power of the investigative agency.This paper has studied the construction of the knowledge Graph in the field of video detection,and completed the following three parts:1)Constructing the data layer of the knowledge Graph of video detection.Based on the data source,extracting the entities in the text corpus by using the BiLSTM-CRF named entity recognition joint model,extracting the semantic relationship of entity based on the dependency parsing analysis,extracting the attributes of entity by using the part-of-speech sequence of the feature words;Realizing the mapping of structured data to the triples with the help of D2R,which lead to the form of the knowledge base of video detection,and linking the entities extracted from the cases into the knowledge base;Managing the extracted and merged knowledge in a dictionary manner;Storing the structured knowledge of video detection in the graph database Neo4j.2)Constructing the pattern layer of the video detection knowledge Graph.By summarizing the video detection knowledge,sorting out five elements of the ontology of video detection{C,A~C,R,X,I},and elaborating in detail the construction method of ontology of video detection.The Protégéis used to generate the ontology model of video detection,including the editing of concept class,object attribute and data attribute.Based on the constructed ontology model,studying preliminarily the mode of knowledge reasoning.3)Implementing and applying the knowledge Graph of video detection.Through the configuration of Deepfinder platform,realizing the visualization of knowledge Graph and exploring the update method of knowledge Graph of video detection.Proposing the application process and thought of knowledge Graph in video detection work,and illustrating the practice of knowledge Graph in video detection according to the video detection case.