Research On Lane Line Detection And Traffic Sign Recognition Based On Machine Vision

With the development of national economy and the continuous growth of car ownership in China,the road traffic environment is becoming more and more complicated,and the problem of road traffic safety is becoming more and more prominent.In order to effectively relieve traffic pressure and reduce the occurrence of traffic accidents,assisted driving technology has become a research focus of intelligent vehicles.The automobile safety driving assistance system is composed of many parts.This thesis mainly studies the two key technologies which are lane line detection and traffic sign recognition.The specific work is as follows:1 For lane line in road traffic scene is easy affected by changes in light,respectively convert the image from RGB to HSV、HLS,and combine the Sobel operator and global threshold method to enhance the gradient features of image.Finally obtain the binary images.2 In view of the problem that the traditional Hough transform method can’t detect the curved lane,use two-way window and second-order polynomial to find and fit the feature points.Firstly,by observing the pixel distribution histogram of the binary image,it can be seen that the peak points of the pixels in the left and right parts of the histogram correspond to the starting points of the left and right lane lines respectively.Then draw the rectangular windows which bottom edge’s middle points are the starting point of left and right lane lines respectively.Then calculate the average value of the horizontal coordinates of the pixels in the window,and take the intersection of the column where the average value is located and the upper edge of the rectangular window as the midpoint of the lower edge of the next rectangular window.Next repeat the search until the rectangular window reaches the upper edge of the image.Finally,fit the lane pixel feature points with second-order polynomial.3 For traffic sign recognition,firstly,preprocess the data set of the training network with size normalization and histogram enhancement;Then train the improved LeNet-5 convolutional neural network by the pre-processed GTSRB data set.The results show that the accuracy of network can be improved to over 97.5%,and the accuracy of test can be improved to 95.3%.4 Verify the accuracy,applicability and anti-interference of the lane line detection,as well as the applicability and accuracy of traffic sign recognition by using the collected road images.