Research On The Recognition Of Traffic Signs And Marking For Autonomous Vehicles

With the continuous development of China's economy,the continuous progress of China's science and technology,the increasing improvement of people's living standards,new requirements for the safety and intelligence of automobiles have been put forward with the vigorous development of the automotive industry.In this situation,as a solution to this problem,autonomous vehicles have brought opportunities and challenges to the entire industry.The key part of the autonomous vehicles technology is the recognition of traffic signs and markings.This paper first studied the current image processing technology and then designed a complete set of traffic sign recognition program,locating and classifying forty-three different types of traffic signs in the database image.It carried out the threshold processing to the pre-processed images and then completed the coarse positioning through the relevant image morphological operations.Based on the research on the characteristics of traffic signs,it put forward six feature descriptors for the further screening of the candidate regions.It also proposed to combine he HOG feature and the Haar-like feature to be the input vector.It trained the support vector machine to recognize whether the current image contains traffic signs or not.Then it classified the traffic signs through the convolutional neural network and obtained a classification accuracy over 80%.This paper studied the recognition algorithm of Hough Transform for the traffic markings and proposed an improved random sample consensus to recognize the traffic signs which tracked each the area where each traffic marking was and limited the parameters of the fitting model.Then it designed the experimental program for verification.Finally,the experimental results showed that compared with the Hough Transform algorithm,the recognition algorithm had the higher accuracy and recognition speed.