Research On Lane Detection And Tracking Algorithm Based On Monocular Vision

With the continuous increase in the number of traffic accidents in China,the research on driving assistance systems for automobiles has gradually become a hot topic.The lane departure warning function is a basic and important module in the driving assistance system of the car.Its core algorithm lies in the detection of the lane.The detection of the lane mainly completes the recognition of the lane from the image and determines the position of the lane with respect to the vehicle in order to monitor the real-time conditions of the vehicle's travel.When the vehicle is unconsciously deviated,it can remind the driver to adjust the state of the vehicle in time so as to avoid traffic accidents.This paper studies the algorithm of lane detection using a monocular camera and proposes a new lane detection algorithm based on machine learning.The main work is as follows:1.A method for automatically determining the area of interest is proposed.The method uses the pinhole model of the camera itself and combines the calibration information of the camera.By calculating the intersection points of the two straight lines formed by the four calibration points in the image,the region of interest can be automatically divided.Since the detection area of the lane is reduced,the time required for lane detection can be reduced to some extent.2.A lane detection algorithm based on machine learning is proposed.The algorithm first uses the calibrated camera installed in the vehicle to capture the front road surface information,and performs the HOG+SVM classification judgment on the small tiles already partitioned in the region of interest.Next,in the picture block that the classifier considers to contain a lane,it used two points to represent the lane section in the small block.In order to ignore some inevitable road background interference,it adopted the Random Sample Consensus(RANSAC)algorithm.Then,for all classified small blocks containing lanes,a parabolic geometry model is used to perform an overall fit.Finally,with proper tracking methods,it can reduce the number of small picture blocks that need to be detected in order to improve the detection speed,and then use the continuous relationship between the previous and the next frames to determine the credibility of the current frame lane detection results.Experimental results show that this algorithm can effectively detect lanes in real time.3.A lane departure warning model adapted to the lane detection algorithm presented in this paper is designed.This model can provide lane warning results in real-time in real-time based on the lane detection information and camera calibration information of this algorithm.