Transponder Recognition System Based On Direction Gradient Histogram And SVM

In recent years,railway transportation has dominated the world’s economic development.In terms of transportation and tourism,people are increasingly inclined to such convenient and economical means of transport;in terms of transportation,railway transportation also plays a decisive role.Various countries are paying more and more attention to the research of rail vehicles,and the speed of trains is constantly being refreshed.At present,the fastest train in the world is the TGV that has been sent,setting a world record at 574.6 kilometers per hour.With the continuous refresh of train speeds,railway safety issues are gradually being taken seriously.As an important part of the railway,the transponder can provide the road information for the train-controlled vehicle-mounted equipment,such as line speed,line slope,track circuit parameters,temporary speed limit,etc.The absence or damage of the transponder may affect the smooth travel of the train and even cause serious consequences.However,the current transponder detection method is manual patrol detection.Obviously,this traditional detection method cannot meet the needs of today’s railway development.Based on this background,combined with computer vision technology,this paper proposes an automatic identification system for railroad transponders based on HOG features and SVM classification.This system has such good robustness to light and shadow effects and graphic translation,and the detection speed also meets real-time monitoring requirements..The system first uses the Gamma correction method in the image processing part to enhance image contrast and correct areas with too high or too low gray levels;the purpose of ROI extraction of the original image is to reduce the complexity of HOG feature calculation as much as possible;based on rails The effect of light and shadow during image acquisition uses histogram equalization in image enhancement technology to highlight the features of the sleepers and make the image clearer.Image binarization is used to select appropriate thresholds to segment the sleeper and non-sleeper regions,and the sub-images of the sleeper region are used instead of the original image as the input image for classification prediction.In the feature extraction stage,the HOG feature vector is used as the image descriptor and the HOG feature Describes the local texture features of the image,which can effectively suppress the influence of factors such as illumination and deformation on the prediction result;in the classification prediction stage,an SVM classifier is used.Finally,use Visual Studio 2017 as a development environment to implement software system development and testing.