The Research Of Real-time Recognition Algorithm About City Traffic Signs

As with the continuous development of urban construction，the popularity of cars, asubstantial increase of motor vehicles and people on the go, made it more congestion thanever, so as the traffic accidents. The increasingly prominent of transportation safety andefficiency of road transport made existing road network through traffic capacity can not meetthe needs of rapidly growing. Thanks for the ITS (Intelligent Transportation System), which isa set of communication, detection, control and computer technology in an integratedinformation system, make it meaningful and has high economic value in protecttransportation and promote the development of the national economy.TSR (Traffic Sign Recognition) is an important part of ITS. In order to maintain asmooth traffic and prevent accidents, it collect and identify the information that containstraffic signs and make instructions or warnings to the driver in time, or control the vehicleoperation directly. Therefore, the core of the research is to improve the recognition rate oftraffic signs, which requires recognition algorithm has a greater robustness that can affect theambient lighting conditions, the vehicle vibration and shelter materials, etc. and be able tomeet real-time requirements.The main contents of this paper include: use simulation software to process traffic signimages in high speed and accurately. The main work is to research the algorithm about trafficsigns determination, segmentation and recognition in-depth.(1) In order to detect traffic signsfrom the captured image quickly and accurately, the paper first divided the possible target areaby using color and adaptive threshold; then filtered the area based on morphology and size,remove these that may not like a traffic sign, finally get the square area containing trafficsigns.(2) In order to overcome the influence of light, vibration and rotation of the vehicle andother factors, we use SIFT algorithm to extract features, for it is invariance to translation,rotation, scale zoom, brightness change, occlusion and noise, and also has a certain degree ofstability in visual changes and affine transformation. To meet the needs of real-time, wechange the original algorithm in a parallel execute way, the simulation results show that the improvements can increase computing speed, and the effect is more ideal.(3) In the phase of traffic signs recognition, the Kd-tree nearest neighbor method and thenearest neighbor ratio are combined to match the feature points, we have write down the timethe system used in each stage, by comparing the data, it can been seen that the experimentalresults shows The program use to recognizes traffic signs are very effective.