Road Speed Limit Sign Recognition Method Based On Evolvable Hardware

Traffic sign recognition (TSR) is a technology which can recognize road signsautomatically. TSR technology can supply drivers important road information andeffectively help the drivers drive safely. The road speed limit sign is the most commonsign that can be seen on the road. They limit the maximum speed of vehicles. Carelessdrivers are easy to ignore these speed limit signs and drive too fast. It always causesserious accidents. The research of recognizing speed limit signs automatically can be avaluable way to provide drivers more road information and avoid potential accidents.In order to solve the limitations of the traditional recognition methods, i.e. longsystem learning and recognition time, and unreadable learning results etc, a virtualreconfigurable architecture (VRA) evolvable hardware (EHW)-based road speed limitsign recognition method was proposed. An incremental evolution strategy and a multipleclassifiers integration method were introduced to optimize the classifier. An EHW-basedclassifier was realized for real-time recognition of four kinds of road speed limit signs,and the performance of the classifier was tested in the real road environment and on ascaled-down autonomous vehicles respectively. At first, through the location and featureextraction processes of the four kinds of traffic signs,90preprocessed feature vectorswere employed in102pictures. The90vectors was divided into two parts: the trainingdataset and the test dataset. And then the EHW-based recognition system was designedand realized, the classifier was trained by the training dataset and tested by test dataset. Inorder to improve the system learning speed and the recognition accuracy, an incrementalevolution strategy and a multiple classifiers integration method were introduced. Theoptimization performance of the EHW recognition system was analyzed and discussed invarious experimental settings. And at last, the recognition system is tested on scaled-downautonomous vehicles. The image distortion problem and real-time recognition problemare the two points that are discussed in the paper. The results show that the proposedscheme is an efficient tool for real-time speed limit sign recognition.