Research On Unstructured Road Driving Area Detection In Complex Scene

With the rapid development of computer vision technology,vision-based intelligent car navigation systems have received more and more attention and investment.In actual work,unstructured road detection will encounter various complex road scenes,such as various road surface coverings,different surrounding environments,and changes in lighting conditions,which increase the difficulty of detection based on vision technology.Road detection is the basis for the realization of intelligent vehicle navigation,especially for the detection of unstructured roads.Only when the drivable area of the road is detected quickly and accurately can real automatic driving be realized.Therefore,it is of great significance to study unstructured road detection under complex road scenes.This paper researches on unstructured roads detection in various complex road scenes,with the goal of improving the versatility,accuracy,real-time and robustness of road detection methods.On the basis of detecting the vanishing points of roads,gradually realize the detection of the drivable area of the road.The main work of this paper is as follows:(1)Research on the detection of road vanishing points.This paper divides the road vanishing point detection into three parts: texture principal direction estimation,effective voting point selection and vanishing point voting method.Use the texture direction of pixels to select effective voting points.In the part of the vanishing point voting method,the relationship between voting points and candidate vanishing points is considered from multiple angles,a distance-weighted local soft voting method is proposed to realize road vanishing point detection under various complex road scenes.(2)Research on road area detection.This paper takes the detected vanishing point as the road constraint condition,and quickly and accurately extracts a triangular area for vehicles to travel.First,for structured roads,based on the classic Hough transform,a Hough transform road detection method that has passed the vanishing point is proposed.Then for unstructured cross-country roads,according to the characteristic that the road direction is basically consistent with the texture direction of most pixels,a road detection method based on texture direction is proposed.Finally,for unstructured rural roads,a road detection method based on local feature fusion is proposed,which uses multi-feature fusion and vanishing point update to achieve road area detection under various complex road scenes.(3)Research on the detection of road potholes.Based on the obvious differences between potholes and surrounding roads,this paper first uses traditional image processing methods such as edge detection and clustering to detect road potholes.Then,the method based on deep learning is used to detect the potholes on the road,and the U-Net convolutional neural network is used to train the road pothole images to obtain the road pothole detection model,which realizes the accurate positioning of the road potholes.In this paper,the driving area detection for unstructured roads in various complex road scenes are realized.The research results show that the method in this paper can detect the drivable area of the road more accurately and meet the requirements of this subject.