Unstructured Road Area Detection And Type Recognition Based On Machine Vision

The environmental perception of road conditions is an important part of auto-driving technology.The road condition information obtained in advance can provide real-time operation judgment basis for the start,turn,and stop operations of the autonomous vehicle during the driving process.This paper takes unstructured roads as the research object,and provides reliable input information for various vehicle control systems by exploring its road area detection and road recognition algorithms,so as to achieve precise control of vehicles driving on unstructured roads.The research on unstructured road detection and road recognition is mainly based on visual information method,which requires the installation of sensors such as cameras on the car.Relying on the National Defense Science and Technology Innovation Special Zone Project,this paper has carried out research on unstructured road area detection and type recognition based on machine vision.Firstly,choose a variety of typical unstructured roads,build a data collection system to collect road images and vehicle speed information,and preprocess the road images;Secondly,use the method combining road features and road models to fit the road boundary;Then extract the texture features of the road images of interest,and design an unstructured road classifier based on Support Vector Machine;Finally,analyze and post-process the road classification results to improve the classification accuracy.The main research contents are as follows:(1)Data acquisition and image preprocessing.Build an on-board data acquisition platform system consisting of test vehicle,camera,USB CAN,controller and computer,select safe and feasible test sites,and use the platform to simultaneously collect road images and speed information of cars driving on a variety of unstructured roads Signals,etc.Perform reasonable preprocessing operations on the collected road images to obtain high-quality images and establish a comprehensive and reliable database.(2)Research on road area detection of unstructured road images.Find the disappearing line of the road to extract the area of interest by using the change of the gray level of the image line.Use the threshold segmentation to divide the road area and the non-road area preliminarily.Then Use connected domain analysis to extract the largest connected areas and remove small areas of non-road connected areas to correct the road area.Finally,extract the road boundary points and fit the road boundary.(3)Research on unstructured road classification based on SVM.Construct training set samples and test set samples of asphalt pavement,brick pavement,soil pavement,sand-gravel pavement by extracting regions of interest from the collected unstructured road images for the classifier;extract texture features and use MATLAB/LIBSVM tools to build Support Vector Machine road multi-classifier,the results show that the road classification algorithm designed in this paper has a higher accuracy.(4)Analysis and post-processing of road classification resultsUse continuous road images based on time and space to test road classifiers based on Support Vector Machine and analyze the results.Then design a post-processing method for the classifier combined with the speed information of cars collected synchronously with the road images.Then simulate and verify the method by using the actual vehicle test data of multiple types of roads.The results show that the method has a better recognition effect on two kinds of road abrupt changes.