Recognition Of Road Traffic Signs Based On Image Feature Extraction

The rapidly increasing development of artificial intelligence has brought the fourth indus-trial revolution to this world.With the growing demand for intelligent living,driverless technology in the field of travel gradually shines in people's vision.One of the most critical technologies in driverless technology is the recognition of traffic signs,that is,the image recognition technology.This recognition process generally includes the collection,prepro-cessing,image segmentation,feature extraction and comparison of the traffic signs.So the content of the thesis is relative to the image restoration,image detection and image classifi-cation of the traffic sign figures that we collect on the real road or in the sky.1.Image restoration:In this thesis,the noise removing through filter and nonlinear stretching method in image enhancement is proposed to address the problems such as incompleteness,blurring,and noise that may occur in the collected traffic sign images,which is able to achieve the brightness adjustment and image information highlighting of the entire screen.2.Image detection:The image detection process mainly includes image color and shape segmentation and similar circularity detection.In this thesis,the RGB color space of the image will be converted to the HSV color space to reduce the complexity and time of the operation.Through setting the threshold range of each component of HSV according to the main color of the traffic sign,the objects in the image are going to be segmented,which will be reserved as the material for feature extraction in the next step.As for shape feature,the signs of circle,triangle and rectangle account for a large proportion.In this case,circularity,can be the important basis and measurement standard for detecting image shape,because every shape has different range of circularity.3.Image classification:This step mainly includes the feature extraction and matching of traffic signs based on Hu invariant moments,BP neural network and Mahalanobis distance.After color segmentation and shape segmentation,Hu invariant moments of the detected image are calculated,and matched with the Hu invariant moments of the standard traffic signs trained by BP neural network through Mahalanobis distance,so as to distinguish whether the segmented image belongs to one sign and which type of sign it is.According to color and shape characteristics,the traffic signs can be roughly divided into six types.This classification method greatly simplifies the recognition process,reduces the data set,and improves the recognition rate.And in this thesis,we detect and recognize the more common seen signs,such as prohibition,warning and direction signs.In order to further recognize the concrete types of the signs,Hu invariant moments,BPNN and Mahalanobis distance will be used in the process of image feature extraction and template matching.Fi-nally,GUI of the whole detection and recognition system algorithm is established based on MATLAB,and traffic images taken at different locations and at different times were identi-fied and tested to verify the timeliness and accuracy of the algorithm.