Research On The Method Of Traffic Sign Recognition Based On Machine Learning

The rapid development of economy and technology always changing is constantly changing our daily life,and the rapid c.hanges in the automobile industry make our requirements on the performance and quality of vehicles constantly improve.The increasing number of road vehicles not only brings great convenience to people's travel,but also makes traffic safety,environmental pollution and other problems more and more serious,which makes the assisted driving technology and unmanned driving technology become the research hotspot of domestic and foreign experts and scholars and technology companies.Traffic signs contain a large amount of road information,so as to reduce the driving pressure of drivers,relieve the pressure of road traffic,reduce the probability of traffic accidents,and avoid the losses of personal safety and property.Therefore,a fast and reliable traffic sign recognition system has become a crucial part of the auxiliary driving and unmanned driving systems.However,the road scenes in the real environment are complex and changeable,and the existing traffic sign detection and recognition methods still have room for improvement in real-time and accuracy.This pap er studies traffic sign identification,and the main work is as follows:In order to solve the problem because of fuzzy image and bad illumination,YCrCb color space and adaptive histogram equalization are used to enhance images,which can effectively improve the influence of bad illumination and make the color of the marked area more vivid than before.For the detection of traffic signs,RGB and HSV color space were used to segment the blue,red and yellow color areas by image mask.After binarization,morphological processing was used to filter out noise points for improving the contour information of the signs.The relative areas that might contain traffic signs were segmented by cutting up the geometric signs.HOG feature from the relative region was extracted.GTSRB database was selected as the sample database,and the data was enhanced to improve the uneven distribution of samples.S VM classifier was used to detect whether there were traffic signs in the relative region.The algorithm was verified in the GTSDB test set and the self-built data set,and the detection accuracy reached 96.29%and 95.54%,respectively,with an average detection time of 0.36s for an image.Aiming at the problem of traffic sign recognition,improve the structure of the tradition al LeNet-5 convolutional neural network by adj usting and optimizing the parameters and algorithm in the model for traffic sign recognition.Experiments were carried out in GTSRB and self-built data sets,and the recognition accuracy reached 98.20%and 98.76%,respectively.The average recognition time was 4.01ms for a sign.Compared with other algorithms,the real-time performance and accuracy were improved,but there was still room for improvement from the actual application.