Method Of Graph Topology Representation For Traffic Vehicle Image

With the rapid increase in the number of vehicles,the contradiction between vehicles and roads has become increasingly prominent,which has caused a series of traffic problems such as congestion and accidents.Vehicle is an important carrier of intelligent transportation system.The representation and retrieval of vehicle images has become a research hotspot in the field of intelligent transportation.How to quickly and accurately identify vehicle types,how to efficiently and systematically retrieve similar vehicles,and how to analyze and explore space-time vehicle information contained in traffic images.These pose challenges for the representation and understanding of the technology of the vehicle image.The representation and retrieval of traffic vehicles images have important theoretical research significance and potential application value.This thesis focuses on the topological graph representation method for traffic vehicle images.And the proposed graph model representation method is applied to tasks such as vehicle image recognition and retrieval.The main research contents of this thesis include:(1)We propose a vehicle image representation method based on a weighted structure graph model.Firstly,we extract the set of component locations of the training sample and do a normalized preprocessing to build an initial complete graph model involving the nodes,edges,and attributes of the graph.Then,we optimize graph nodes and graph attributes based on the Fisher method and voting mechanism.Finally,we calculate the contribution of graph attributes to vehicle retrieve and generate a weighted structure graph model.Experimental results show that the proposed graph model method can effectively represent vehicle images.(2)We put forward a vehicle image recognition method based on weighted structure graph model and support vector machine classifier.Firstly,we use the weighted structure graph to represent the vehicle image.Then,use the preferred attributes of the graph to train the support vector machine classifier,and realize the recognition for the VIS2015 vehicle image set constructed in this thesis.Compared with the experimental results of local binary patterns and convolutional neural networks,this method can effectively improve the recognition accuracy and efficiency.(3)We have integrated the equivalent local binary pattern and the weighted structure graph representation model,and achieved vehicle image retrieval.First of all,we use the local binary pattern to extract the texture features of the car emblem image,and use the weighted structure graph model to represent the vehicle image as a topological graph.Then we implement feature weighting and fusion of these two features.Finally,we achieved high recall and precision rate vehicle type retrieval through similarity matching.(4)Based on the Neo4 j graph database,we build a spatio-temporal image representation and path retrieval framework.We establish a time-space traffic image representation model to realize the storage and presentation of traffic image data containing time and space information.We propose a space-time constraint rule and a vehicle path retrieval method,which is based on the spatio-temporal representation model,and realize the trajectory retrieval according to time zones,regions,and vehicle types.