Research On The Expressway Network Video Information Support System(ENVISS) And Related Key Algorithms

Nowadays,with the incremental development of expressway network in China,the construction of infrastructure for the three traditional systems including communication,charge,and monitoring is turning to information intellectualization.In this background,Intelligent Transportation System(ITS)is growing increasingly,and the integrated application systems reflecting road network information are emerging.The video resource,as a key role in information construction of ITS,has attracted extensive attention because it is visualized,effective and information-rich.However,there are many issues on how to manage and apply the video resources for expressway network in current phase.The issues include:it is difficult for information sharing and exchange in the existing heterogeneous video application systems;different application systems need to retrieve and analyze the video resources by themselves,which results in duplication of work and inefficiency;the video images from different resources vary greatly in quality due to the impact by network transmission performance,or different class of camera equipments,or weather condition in field.So it is unable to provide the real-time report effectively through manual work and subjective feeling;the video resources obtained along the express way are relevant.For example,the visibility and traffic jams are distributed dynamically and regionally.It is hard to make accurate information release and take corresponding actions if we have no idea about the regional distributed resources and the cause behind.Hence,it is vital to build the Expressway Network Video Information Support System(ENVISS),with researching on two major areas including the system architecture and the related algorithms for video information processing.This paper first provides an overall introduction of the research on traffic video information system both locally and overseas.Then it describes the three-dimensional layered network architecture,LDM~3,with self-owned intellectual property right.Then ENVISS built up based on LDM~3 is proposed and introduced in details:in the system,to improve the traditional TCP/IP architecture for ITS,the information fusion layer is newly introduced that is on the top and in parallel with application layer.The new architecture is effective and efficient for information exchange and has good compatibility with the existing system;in existing way,the different application systems retrieve and process the raw video data independently.ENVISS changes the legacy way by collecting,mining and refining the video data accordingly to the needs of application systems,then pushing the information proactively and automatically;it separates the preliminary processing for pixel level and fine processing for information level for video resource,and unifies the preliminary processing as it is complicated with great computational complexity,which improves the efficiency of the system significantly;it achieves the intelligent management of video resources,so that the administrator can keep track of the real-time report of the video quality,then take corresponding actions timely to ensure the health of the system.In addition,the related algorithms for video processing utilized in ENVISS are introduced in sequence:Regarding image quality evaluation,two algorithms are proposed and researched in this paper,including the color image quality evaluation method based on Human Visual System(HVS)and quaternion,as well as the image quality measurement method for blind image restoration.The simulation results demonstrate that the proposed method has a higher consistency with subjective evaluation than others.For image restoration and enhancement,a semi-blind image deconvolution algorithm with spatially adaptive total variation(SATV)regularization is introduced.This algorithm reduces the noise in flat regions and preserves the edge information for defocused or motion blurred images.In addition,this paper adopts a defogging algorithm,which converts the image to transmissivity triple-channel with low correlation and make adaptive enhancement based on Logarithmic Image Processing LIP model.The algorithm significantly improves the visibility of image.At the same time,the issue that the tone of image is too dark after defogging is mitigated.In the area of image data mining,this paper researches on the vehicle feature extraction and classification based on machine learning and pattern recognition.It details the method of background modeling based on Gauss model and feature extraction based on sparse coding.Also,the performance of the method is evaluated with contrast experiments by using the raw image retrieved from surveillance video.The result shows that the algorithm proposed improves the recognition accuracy comparing with the traditional recognition methods.What's more,this algorithm leads to a satisfactory recognition result even the video image is with low resolution or the target is covered by other objects.Based on the existing macroscopic traffic situation system with map interface,this paper builds a virtual reality system to demonstrate the microscopic traffic situation for hotspots.With the input of data for vehicle model,speed and so on through video detection,the system can simulate the traffic situation in an intuitive and friendly manner.By adopting the enhanced cellular automation microscopic traffic model,this system can be used to provide predictive data for blind areas like tunnels,bridges etc.In a summary,this paper researches on ENVISS as well as the related key algorithms,which makes significant contribution for the provincial level projects like the Demonstration Project for Intelligent Traffic Cloud Service Platform for Jiangsu Province,and the Demonstration System for Traffic Information Release and Decision Assistant Based on Information Detection,Mining and Fusion.At the same time,it solids the foundation for the implementation of the Informational Demonstration Project for Ninghuai Expressway.The major innovative areas of this paper include:·Proposing ENVISS based on LDM~3.This solution enriches the content of information fusion layer and defines the specification for the interface with big data.In addition,it proposes the separation of the preliminary processing for pixel level and fine processing for information level,and changes the legacy passive way to a proactive way for utilization of video resources.The benefits include:it has good compatibility with the existing heterogeneous systems so that the resources can be shared easily;it reduces the duplicate work on complex computation and balances the load between the servers and the cloud,thus the effectiveness and efficiency can be improved significantly.·Optimizing the algorithms for image quality evaluation.By applying the theory of HVS,quaternion and Total bounded variation(TBV),the image quality measurement methods have been enhanced with more accurate subjective evaluation than others.Therefore,the effective and timely online management of video resources can be realized.·Optimizing the algorithm for vehicle classification based on sparse coding.This algorithm is an effective solution when the image is with low resolution or the target is covered by other objects.Thus the accuracy and anti-interference for subsequent traffic flow parameter detection can be significantly improved.·Proposing and implementing the virtual reality system to demonstrate the microscopic traffic situation for hotspots,combining with the macroscopic traffic situation in road network.This system provides a demonstration platform for public services and road supervision,with a convenient and friendly interface for multi-level,multi-scale and multi-dimension.