The Automatic Analysis Technology Research On Moving Objects Of Intelligent Transportation System

Intelligent Transportation System (ITS) is a promising intelligent system providingreal-time service for traffic management. With the combination of multiple fields, the ITSconsists of the acquisition, processing, storage and distribution of information. With thedevelopment of video monitoring systems and the advances in video image processingtechniques, application research based on video image processing of ITS has beenwidespread and deepgoing. It is of great practical value and theoretic importance to performan better research on video image processing algorithms of traffic video in order to make theITS more intelligent and functional in handling problems such as accident judgment, vehicleclassification, traffic flow parameters detection and moving target tracking.With the moving targets in traffic video as the main object, this thesis focuses on theresearch of video and image processing algorithms. The content is comprised of detection,extraction, identification and tracking of moving targets, as well as calculation andidentification of the traffic flow parameters, providing support for the intelligentizing ofITS. The primary contents are as follows:The moving targets were extracted with motion detection method and imagesegmentation method combined. First moving targets detection based on the OrthogonalGaussian Hermite Moments were introduced for the complexity of traffic video. Then imagesegmentation of a single frame from the traffic video was performed with Markov model. Inorder to reduce the time complexity of the algorithm, Graph Cuts method was adopted. Seedpixel extension method was proposed and mattes of the moving targets were generated basedon the two aforementioned methods, thus realizing automatic extraction.An automatic moving target extraction method based on auto-generated Scribbles wasproposed by combining Orthogonal Gaussian Hermite Moments, morphology processing,distance transformation and threshold segmentation. Performance of method is superior tothe method proposed in the chapter above for it depicted both the foreground and the background. Through experiments, comparisons were made between Graph Cuts and Close-formSolution algorithms, and the influence of the parameters during Sciribble auto-generationwas analyzed, providing conditions for further processing. The target recognition and matching based on adaptive multiple features algorithm wereproposed. Color, edge and moment were selected as single feature in describing the targetsfirst, and HSI Color Histogram, Edge Orientation Histogram and HU invariant moments ofmoving targets were extracted. Then image distance was selected as the similaritymeasurement and random region weighted estimate was proposed as the exploration policy,where the weight of each single feature could be altered according to the result of singlefeature matching.Following the theory of partical filtering theory, according to the methods in last chapter,with a second order constant model as the state model and an adaptive fusion multi-featuresmodel as the observation model, a moving target tracking method was proposed. Thevariance and weight of the likelihood functions of each single feature in observation modelcould be real-time updated according to the tracking results of previous frame, thus resultingin an adaptive observation model and increasing the accuracy of moving targets tracking.A traffic flow test platform was constructed, and the parameters of the traffic flow weredetected, thus realizing the transition from theory to practice and providing basis forimproving the capability of traffic video monitoring.