Traffic Surveillance Video Retrieval Based On Semantic Content

The extensive use of traffic surveillance systems has reduced the number of traffic accidents to a certain extent,and also brought a lot of convenience to traffic management departments in managing traffic.However,it is extremely difficult for people to extract the required information from massive surveillance video data.In most cases,manual intervention is required,which takes a lot of time and human resources.For example,the current traffic management department uses a wider range of traffic surveillance systems to capture traffic violations,and manual intervention is still required to accurately determine whether a driver is do something illegally.This article presents the solution to the following issues.(1)For humans,we are more accustomed to using keywords-based search to retrieve the required content,but the current keyword-based video retrieval system still has many limitations.When non-professional users query video-related content,it is difficult for them to input accurate and comprehensive coverage of search terms,resulting in the query results are limited.And it can't meet the user's initial query intent.In order to get a result that better meets the user's query requirements,this paper introduces a domain ontology for a specific application domain,semantically expands the query input statement,and thus obtains a query result that more closely matches the user's needs.(2)In view of the fact that there is lack of more comprehensive ontologies for traffic surveillance events,this article creates an ontology based on traffic surveillance events and verifies the rationality of the ontology through logical reasoning and actual events.(3)In view of the fact that existing video retrieval schemes cannot meet the need to directly locate specific traffic accident scenes,this paper applies image match video retrieval to retrieval of specific traffic accident scenes.For traffic accidents with obvious characteristics,the system has a High accuracy.