Traffic Elements Perceiving And Event Semantic Understanding Based On Video Image

At present, China will actively explore the applying new technologies to improve therunning of the transportation system and efficient transport management. Intelligenttransportation system (ITS) is one of them. It not only makes transportation become moreefficient, but also make our means of information exchange more convenient. Simultaneouslytime, incident understanding based on the video image of the traffic is being more and moreconcerned. But traffic information collection technologies based on video image processingtechnology have yet to be further improved and perfected; the traffic incident understandingmodel and identification method have yet to be further improved. This article is based on theabove the background, do further research about perception of the traffic elements and eventsemantic understanding about the traffic event.First, according to the methods of semantic understanding having a strong dependenceon the type of event, do further research about versatility and the generality, the main contentsare: analyzing of the model of the traffic incident from the perspective of the framework;the characteristics of the underlying video is divided into two types: static characteristics anddynamic characteristics; given the traffic incident semantic expressions according to the logicof the rules of natural constraint language (Natural Constraint Language, NCL), while givingoverall process framework to achieve semantic understanding of the traffic incident.Second, according to the important factors of accuracy and real-time performance cannot meet the requirement simultaneously in current vehicle recognition, a compound imagematching model and a recognition method were developed, namely classify vehicles firstly byutilizing Harris corner, then classify vehicles detailed by using SIFT feature. The methodshortens the time under guaranteeing the accuracy to be kept essentially constant, real-timeperformance was improved greatly.After various low-level features gotten, then the order of traffic event recognition wasthat semantic unit, basic event, high-level semantic event. Finally, high-level semantic eventrecognition utilized Hidden Markov Model (HMM) model to get high-level semantics,diminishing the semantic gap to a certain extent. Because the traditional method torecognizing traffic events in complex scenes is the lack of real-time performance, improved the process of understanding traffic behavior, the experiment shows its effectiveness.