| Intelligent transportation is the trend of future transportation development,and automatic driving has also set off a boom in recently.Automatic driving is often inseparable from machine vision.Image recognition using machine vision can quickly collect road information and feed it back to neural network for such operations as road monitoring,automatic driving,etc.It is understood that the number of accidents in the world every year is still very large,and the development of intelligent transportation can also reduce the probability of accidents to a certain extent.This paper takes the detection of road traffic markings as the main task,and carries out the detection of zebra-crossing,conventional guide line,lane line and vehicle line pressing through the combination of image processing and deep learning.The specific work contents are as follows:(1)In the zebra-crossing detection of road traffic markings detection,a zebra-crossing detection method based on cascaded Hough transform and vanishing point characteristics is proposed.This method applies the principle of parallel coordinate system for detecting straight lines to zebra-crossing detection,and represents each edge point of the image obtained from edge detection in the parallel coordinate system to find image lines and the vanishing point.And the vanishing point coordinates are used to exclude the interfering straight lines,and finally the zebra line fitting line segment is obtained by image lines error correction and edge points fitting.The experimental results show that the detection of straight lines and vanishing points using CHT has significant advantages over HT in terms of detection accuracy and detection speed.The vanishing point can also be used to exclude interfering straight lines and provide support for the accuracy of zebra-crossing detection.In addition,this method does not need to use region of interest extraction to obtain the zebra-crossing location information more accurately,which also provides a reference method for road detection in some specific situations.(2)In the road guide lines detection of road traffic markings detection,a road guide lines detection based on the improved YOLOv5 algorithm model is proposed.The method first greatly increases the training samples of the dataset through techniques such as image enhancement,and improves the effectiveness of the dataset training by modifying parameters such as anchors of the YOLOv5 algorithm model.At the same time,a CBAM lightweight attention module is added to the model side to further improve the model’s extraction capability for small target objects.The experimental results show that the method can successfully identify the road guide lines.And by comparing with the current mainstream target detection models,it is found that the optimized model has improved in both recognition accuracy and recognition speed.(3)In the straight lane vehicle line pressing detection of road traffic markings detection,A new method for vehicle line pressing detection is proposed based on the theory and experiments related to Chapter 2 and Chapter 3.The method first uses the MOG2 background modeling algorithm for automatic road background acquisition,and then adds the instance segmentation algorithm and the Bi FPN weighted bidirectional pyramid structure to the YOLOv5 model to enable it to obtain the vehicle edge information after detecting its position and to improve its detection accuracy.A new optimal judgment condition for vehicle line pressing is also proposed so that it can eventually perform the judgment of vehicle line pressing while detecting lane lines and vehicles.The experimental results show that the method can successfully achieve the detection of vehicle line pressing,and compared with the instance segmentation plus lane line detection algorithm Mask R-CNN+Lane Net,it has improved the lane line recognition accuracy,vehicle recognition accuracy and inference speed. |
