Research On Lane Determination Technology Based On Vanishing Point Estimation And Extreme Value Filtering

Since the new century,with the progress of science and technology,the role and function of the car have undergone a revolutionary change.The intelligent development of the car began to lead a number of industries to change.At this time,both domestic and foreign Internet companies,national auto companies,and universities do the research on the unmanned vehicles and advanced driver assistance system(ADAS),which can be seen the development of unmanned vehicle technology and advanced driver assistance system has a considerable prospect.The ADAS as a vehicle driving safety auxiliary system,can effectively avoid traffic accidents caused by driver's distraction during the driving.In this case,the main cause of traffic accidents is the car speeding,crossing the road,drink driving and fatigue driving and so on.Therefore,it has a positive meaning to reduce the driver's artificial mistakes to decrease the incidence of traffic accidents.Lane line detection in the ADAS plays a very critical role,it can be used for vehicle navigation,lateral control,anti-collision or lane departure warning system.The current mainstream approach is used by the machine vision method for lane line detection.Generally,we can classify it into three broad categories:recognition algorithm based on lane mark gray feature,recognition algorithm based on lane mark feature and recognition algorithm based on road model.However,based on the above-mentioned algorithms,we 'find that some methods will fail or cause the lane line to be inaccurate when there are some abnormalities such as the change of the road structure,the uneven illumination and the sudden turning or changing of the road.After analysis,we believe that most of the reasons for the situation are due to the image preprocessing when the segmentation effect is not so good,resulting in the subsequent lane mark line detection and identification of deviation.Based on the analysis and research of driving support system,this paper focuses on the detection of lane mark in the problem of lane inspection in structured road image.In order to solve this problem,vision-based detection methods are widely used,usually focused on edge information detection.However,the use of only one edge detection operator in a variety conditions of the road will lead to low detection accuracy and detection error.This paper proposes a method for combining vanishing point estimation with multiple edge detection of lane mark operators.The proposed method not only improves the accuracy of lane mark detection,but also has a good robustness for the areas which have obvious changes in light and shade.In addition,we introduce a filtering method base on adjacent areas that is called extreme filtering.Experiments results show that the method can improve the detection rate and accuracy of lane mark detection.