Lane Detection Algorithm Research On Structure Road Based On Vision Sensor

At present,with the increase of car ownership more traffic accidents are brought.The high incidence of traffic accidents to allow the government to issue more laws and regulations,forcing the automotive industry continues to increase investment in technological innovation.Now,the researches about the security technology are mainly focused on active safety,passive safety,intelligent driving assistance system Intelligent Connected Vehicle and ect.Intelligent system can make the vehicle steering,braking or accelerating according to the information about the surroundings from the sensor on the car.Such a series of actives are based on the right perception of the surroundings.The extraction of lane line is a very important part in the perception of surroundings,accurate and real-time lane detection is a key and difficult point in this field.This paper mainly focus on the lane identification of structured road,according to the characteristics of the structured road and its complicated environment,some solutions are put forward from the aspects of the real time of the algorithm,the accuracy of detection and the adaptability to the environment.In this paper,the main research methods and research contents are shown in the following aspects:(1)To collect the relevant papers about lane identification,and analyze the research status of lane detection algorithm and lane tracking algorithm.Learn the advantages and disadvantages of various sensors and,and find the problems in lane recognition algorithm.(2)Study the Inverse Perspective Mapping(IPM)and camera calibration based on camera imaging principle.According to the imaging principle of the pinhole camera,the obtained image is converted to an aerial view by IPM.In order to ensure the feasibility of inverse transform and transform get the vanished point coordinates guarantee the quality of the sampled image,the distortion of the lens is calibrated by the template method.(3)The Bessel curve Lane fitting based on random sampling consensus algorithm is studied.In the image preprocessing stage,the Gauss filter is used to improve the signal to noise ratio(SNR)of the parallel vehicle moving direction,meanwhile it can effectively reduce the amount of calculation.Thresholding method is introduced to binaryzation which can separate the lane pixels from the non-lane pixels.Through binaryzation we can find the region of the suspected lane and use RANSAC to fit the three order Bezier spline in the region,evaluate the length of the fitting lane line and bending degree.(4)Sduty combines lane tracking algorithm with Kalman filter noise analysis.Kalman filter tracking algorithm is used to track the 4 control points of the Bessel curve.For more accurate tracking,the Kalman filter process noise and measurement noise are extracted by experimental method.In the end,algorithm is evaluated through the method of experiment and contrast.To evaluate the algorithm,use the real-time lane data collected by a real car driving on the structured road.In addition,the algorithm and other algorithms are evaluated under the same evaluation criteria.The evaluation results show:(1)After the inverse perspective transform the image using the Gauss filter and the local threshold processing method makes the image has a higher signal to noise ratio.(2)Kalman filter lane tracking algorithm can track the real-time frame rate of 29.1 frames per second,under various conditions,the average tracking accuracy can reach up to 86.8%.