| Along with everything in the society in a rapidly changing,and people for the requirement of standard of living is increasing.Under this background,the car has gradually become the families cannot leave the means of transport.The popularity of cars can bring convenience to people’s life,however,at the same time,it also led to the increasing range of road traffic accidents and gives a lot of families had brought a nightmare.The diversity of the complexity of road conditions and traffic signal always cause drivers ignore the excessive concentration of attention and omissions or deterrent to all sorts of traffic sign warning,end with tragedy.Automatic recognition of traffic sign,therefore,as the core technology of auxiliary or unmanned driving system arises at the historic moment,and quickly developed,as a more security for car use personnel safety.The current domestic and international research on traffic sign recognition methods emerge in endlessly,including many excellent recognition method.But most of identification methods are based on static traffic sign image,or based on the results by computer terminal’s detection on natural traffic sign,then,end in recognition on computer terminal.Such method often can get good recognition effect,but both of them has low practical efficiency.Only recognition method based on mobile terminal can be applied to practice,to solve the problem of real-time traffic.The system is divided into mobile client and server.Mobile terminal real—time detect traffic signs subgraph and then send it to server for recongnition.On the mobile client designed for sign detection,this thesis adopts color separation and combination of Hough transform to detect the required information.First,we use the HSV color model for color separation on the mobile terminal real-time frame.We extract the sensitive area of the color and eliminate noise area and select candidate area of hough transform shape recognition.We select the candidate area of Hough transform shape recognition,and complete primer detection of traffic sign.Finally,we compress candidate picture to and send it to the server to identify.On the server side designed for sign recognition,this thesis adopts a graph embedding recognition algorithm based on sparse representation.The algorithm to embed the sparse representation of the optimization results and figure identification of together.This algorithm also retains the local relational structure of data collection,and reflects the discriminant information.Traffic signs has particularity.They are divided into a variety of forms,under every kinds of multiple small classes.So we will through the relationship between class and class to hierarchical composition.After the identification,the server send the result to the mobile client,complete the operation of the whole system. |
