Research On Traffic State And Incident Identification Methods For Urban Freeways Based On Cyber-physical Networks

Intelligent transportation monitoring system based on computer vision is one ofmain application measures to alleviate even solve traffic congestion and abnormalevents. But at present, all of main application systems have certain insufficiency onthe mode of internal information transmission and control strategy, making theprocess of transmission not timely, or that the control center can’t update relatedparameters of detectors. On the other hand, in the system, the general trafficparameter extraction algorithms are hard to adapt complicated traffic status. Onceencountering congestion, the validity falls sharply. In additional, traffic staterecognition is mainly on the basis of fixed threshold to estimate the level ofcongestion or divide the traffic state into several classes, both of them neglecting thelarge amount of historical data, which cannot reflect the overall feeling of trafficparticipants and the actual road situation.To solve these problems, taking urban expressway as the research object, acyber-physical system to realize the real-time monitoring of urban expressway,through traffic parameter extraction, traffic incident preliminary automatic warning,traffic state quantitative identification, and incident detection based on videotransmission and feedback control, is designed and implemented. This systemoptimizes the system model and work process, mainly through designing the mode ofpriority to data flow internal information transmission and control strategy of incidentdetection based on abnormal video stream transmission and feedback control. Then,we respectively do researches on the status quo, application design andimplementation on the three main functional algorithms backstage, which are trafficparameter extraction algorithm based on computer vision, traffic incident detectionand traffic state quantitative identification algorithm.In the part of traffic parameter extraction automatically，a real-time extractionmethod, for the parameters such as flow, average speed and time occupancy ratio, isput forward based on the two kinds of time-spatial images (TSI) PVI and EPI, whichis improved by the method based on the TSI. First of all, the PVI and EPI imageacquisition process is improved and optimized, and adaptive binarization methodbased on CM clustering is used to deal with TSIs. Secondly, in view of the situation ofthe same monitoring sections to install multiple cameras, in order to integrate the PVIand EPI binarization image respectively, the method of visual information fusionbased on Probability Fusion Map (PFM) is improved by optimizing camera weightcalculation method. Finally, based on the component notation method, a reasonableand simple traffic parameters extraction method is designed, which is based on thePVI and EPI binary images. In addition, a simulation case is implemented by takingthe road section near Jinsong Bridge in Beijing as an object, and the result shows thetraffic status adaptability and real-time performance of this method.For the part of traffic incident detection, the method of traffic incident detectionbased on spatial and temporal information of single cross section is proposed.According to this method, an algorithm of approximate normal distribution Bayesian decision with minimal risk is proposed to adaptively train thresholds. This methodmakes full use of time and space information, and avoids the omitting of abnormalevents.For the traffic state quantitative recognition part, we propose the method toobtain continuous values based on the3categories of traffic state identification model,which divide traffic to smooth, larger traffic flow or congestion state, and throughestimating the relationships between the current traffic parameters and trafficparameters of the3typical states. Based on the3classification model, this paper putsforward the improved FCM algorithm with gray comprehensive evaluation method.Among them, in the training phase, PSO-FCM algorithm is proposed for unsupervisedtraining cluster center matrix. The algorithm uses the particle swarm algorithm (PSO)adaptive selection fuzzy indices, and considering the weight of each component aboutthe vector of traffic parameter to improve the objective function of FCM algorithm. Inthe real-time identification stage, we adopt the gray comprehensive evaluation method,and the output is a continuum of values within a certain range. We train models basedon large amounts of historical sample data. And then through the simulation andanalysis of continuous time series of traffic parameters, it proved that the proposedmethod can reflect the time gradient trends of transport, and the adaptive ability forroads, the recognition accuracy and effectiveness are all good.Finally, the MFC application software system of traffic surveillance based oncomputer vision is designed and implemented, which consists of2MFC applicationsoftware corresponding the video processor and traffic monitoring client respectively.Taking the expressway video streams near Jinsong Bridge in Beijing as an example, adetailed description of system function and the running process is implemented.Through functions demonstration, the software application system basically realizedall functions of expressway real-time traffic monitoring system.