Research On Lane Departure Warning Model In The Highway Scene

With the continued growth of car's ownership in China,traffic safety has received more and more attention.Gradually,people began seeking to use computer technology to keep driver's safety.However,the environment in the real road scene is always complicated and changeable.So,how to realize the vehicle's departure warning in such scene has become the hotspot and difficulty for the domestic and foreign scholars.In this paper,we put forward a model of LDWS(Lane departure warning system)suitable for the actual situation of China's highway based on machine vision after giving a full study of LDWS in the domestic and foreign.The model mainly includes three modules:road images' pre-processing,lane recognition and tracking,departure warning and decision making.And main work of this paper also revolves around these three modules.(1)Road images' pre-processing: Firstly,in order to remove the redundant part of image(such as the sky,trees,etc.),the paper establishes two ROI(regions of interests)to deal with the left and right lane marking line.Then,the image is processed by RGB Adobe(1998)method.After that,the Pyramid filter is used to implement the road image's enhancement.And in the most important part-edge detection,we chose an improved adaptive canny algorithm to extract the edge of lane,because this algorithm is most suitable for the highway scene.Finally,by setting up a fixed threshold,the binary image of lane is obtained,which greatly simplifies the data quantity in the subsequent process.The result of experiment shows that the image pre-processing method proposed in this paper can greatly simplify the lane image,which provides a good data source for the subsequent lane recognition.(2)Lane recognition and Tracking: To solve the problem of detection model,the paper chose straight line model to detect lane markings given that the requirements of real-time and accuracy has been set.Meanwhile,it also gives error analysis when using this model to detect curved lane.Then,we improve the PPHT(Progressive Probability Hough Transform)algorithm to extract the pixels of lane markings.To solve the problem of double edge,this paper uses the Least Square Method to fit the lane markings,so that the fitting line is located in the central area of lane markings.After that,it proposes an interfering-line-avoid algorithm based on lane markings' weighted length and distance.In the end,we propose an algorithm and a lane tracking mechanism to predict the next ROI area and identify the lane markings based on the previous frame.And the experimentalresults show that the algorithm proposed by this paper can identify the lane markings accurately and timely.(3)Departure warning and decision making: First of all,this paper analyzes the importance of warning time's choice,which must be made based on the consideration of the false alarm rate and driver's reaction time.Then,based on the detailed description and comparison of five commonly used departure warning and decision making algorithms,it chooses an algorithm based on the slope of lane markings to define the vehicle's deviation after pointing out the deficiency of camera calibration method which is mainstream currently.Finally,the experiment's results show that the algorithm is simple and efficient on detection of vehicle's deviation.