Research On Unstructured Road Recognition And Lane Departure Detection In Mining Area Based On Machine Vision

With the implementation of national and local mine intelligent construction standards,driverless trucks in mining areas have become one of the main research contents of intelligent construction of open-pit mine production.Due to the unstructured roads and complex and changeable road environment in open-pit mining areas,automatic detection and identification of driveway of driverless trucks and accurate discrimination of lane deviation are important foundations to ensure the safe driving of driverless trucks,It has become one of the key technologies of unmanned driving in open-pit mining areas.The image recognition method based on deep learning is very suitable for the application and analysis of image data with nonlinear,high-dimensional and complex relationship.This paper makes an in-depth study on unstructured road recognition and lane departure detection of unmanned trucks in mining area based on machine vision method.This paper mainly includes:(1)Construction of unstructured road image data set in mining area.Aiming at the problem that the collected road environment image data of open-pit mining area is single,this study analyzes a variety of image filtering and image enhancement algorithms,and adopts the methods of medium filtering and histogram equalization to preprocess the collected road image of mining area to enhance the characteristics of road image,adopts labelme software to label the image at pixel level.Construct high-quality data sets for road recognition and lane departure detection in subsequent mining areas.(2)Unstructured road recognition in mining area based on improved bilateral segmentation network.In view of the fact that the road area in the open-pit mining area is not obvious and the traditional computer vision method can not quickly and accurately identify the driving area in front of the vehicle,this study adopts a depth learning method based on the improved bilateral segmentation network,uses the depth separable convolution module to improve the lightweight of the detail branch,so as to improve the operation efficiency of the model,and combines the convolution block attention unit to improve the recognition accuracy of the model,Finally,the sampling on the feature map is improved to enhance the performance of road edge details.The experimental results show that the optimized model can achieve better results in the task of fast and accurate recognition of Mining Road area than other classical models.(3)Lane departure detection in mining area.The traditional lane departure detection method is mainly for structured roads and needs complex multi coordinate system annotation,which is not suitable for complex and changeable mining area road detection.This paper proposes a simple and efficient lane departure detection algorithm for unstructured road scene in open-pit mining area.Firstly,the lane recognition image obtained in the previous part is preprocessed to obtain a smooth road edge,and then the least square method suitable for unstructured roads is selected to fit the lane edge line,so as to obtain the slope and intercept of the left and right lane edge lines respectively,and use the image to obtain the lateral deviation parameters,so as to obtain the characteristic parameters of lane deviation,Finally,the BP neural network algorithm optimized by particle swarm optimization is used to train and learn the characteristic parameters.The experimental results show that the model has higher accuracy and better robustness for lane departure detection task in mining area than other classical classification models.Through the research of this paper,the method of combining machine vision and deep learning can better solve the tasks of unstructured road recognition and lane departure detection in mining areas,and provide a certain reference basis for the real scene navigation and safety early warning of driverless trucks,which has important theoretical value and practical significance.