| In recent years,the technology of unmanned vehicle has been a hot research in the transportation system.As traffic problems are increasingly,one of the ways to effectively solve traffic problems is to realize intelligent driving system completely,and the recognition of road traffic markings is a very important technology in intelligent driving system.How to quickly and accurately identify the road traffic markings complex and changeable real scenes from the collected videos is a great valuable subject to research.This paper mainly studies image preprocessing,road traffic marking detection and road traffic marking recognition.The main work is as follows:1.Adopting pre-processing operations such as noise elimination,image enhancement and gray-scale conversion on the image obtained by the frame cutting.In order to reduce the noise existing in the image,the mean filtering,median filtering,Wiener filtering and image wavelet domain filtering were compared experimentally,and the median method with the best effect was selected for noising.In order to solve the problem of low contrast of road traffic markings,the MSRCR method based on color restoration is used for image enhancement processing.Finally,in terms of grayscale conversion,the three logarithmic conversion methods in Open CV are mainly analyzed.Through the comparison experiment,the first logarithmic conversion method is finally used to process the road traffic marking image,which effectively improves the later detection and the speed of recognition.2.In order to be convenient for the recognition of road traffic markings later,it is necessary to detect the road traffic markings in the picture.This paper adopts many detection methods of road traffic markings based on edge detection.For the edge detection,the canny algorithm and the Marr-Hildreth algorithm are studied.Through the experimental comparison,the Marr-Hildreth algorithm is used to detect road markings.Using the mask technology to extract the ROI area in the road traffic image,and the original image is segmented by the minimum rectangular boundary of the contour to realize the segmentation of road traffic lines.3.First,the experiment focus on the ORB algorithm.The algorithm firstly detects FAST feature points on the image,and then it calculates the descriptor of feature point by the Brief algorithm.Lastly,the algorithm compares the collecting images which paired with the feature point with the images which are from the template database to recognize.The experiment shows the ORB algorithm will reach a better accuracy by combining the fast feature point method and the Brief method.Then the application of the hash algorithm in road traffic marking recognition is studied.The algorithm generates a hash value for each image,anddetermines the similarity between the images by comparing the hash values of different images.Based on the study of the mean hash algorithm,the mean is replaced by computing DCT(discrete cosine transform).Experimental results show that the recognition of the perceptual hash algorithm is better than the average hash algorithm.Finally,in order to improve the accuracy of the network model,an improved method is designed on the Alex Net convolutional neural network.This method uses the last three layers of the Alex Net model,and uses the XGBoost classifier that performs better on the classification problem to replace the Alex Net model’s softmax classifier for classification processing.Experiments show that the improved neural network reduces the training time of the original neural network,and improves the accuracy of traffic road marking recognition.Based on the research methods above,combined with the Open CV open source visual library,this paper designs and implements a road traffic marking recognition system based on the MFC framework.The system implements the methods above in sub-modules to achieve visualization. |
