Research On Lane Keeping Control Algorithms Based On Machine Vision

The lane keeping assistance system is an important part of active automotive safety systems,which can effectively prevent drivers from causing traffic accidents caused by lane departures.For the lane keeping assistance system based on machine vision,this thesis studies the algorithms about camera calibration,lane detection and tracking,lane departure warning and lane keeping lateral control,which can be involved in it.In Chapter 2,this thesis studies the algorithms for camera calibration.For cameras’ intrinsic parameters,a calibration scheme based on OpenCV is designed,and the LIBCBDETECT algorithm is used to improve the success rate of detection for checkerboards’ corners.For cameras’ extrinsic parameters,this thesis first detects the optimal vanishing point of the road corresponding to a continuous image sequence,and then uses this vanishing point and the camera installation information to calculate the extrinsic parameters,which can achieve fast and accurate calibration for extrinsic parameters.By the way,this thesis also elaborates the principle of measuring distance use a monocular camera in this chapter.In Chapter 3,this thesis conducts research on the algorithms of lane detection and tracking.In the image preprocessing stage,an image preprocessing scheme based on the linear filtering based on lane-markings’ width feature is designed,and good lane-marking feature maps are obtained with the help of this scheme.In the feature points extraction stage,this thesis detects the starting points of lane-lines at first,and then applies a sliding window searching method with direction guidance to extract feature points of lane-markings in the feature maps,and by this means,good extraction results can be obtained.In the lane model fitting stage,least squares estimation is selected to fit quadratic curve models for lane-lines.In the lane tracking stage,this thesis takes the starting points of lane-lines and the parameters of lane-lines’ models as tracking objects,and designs a reasonable tracking strategy to improve the robustness of the lane detection algorithm.In Chapter 4,this thesis compares the advantages and disadvantages of some existing lane departure warning algorithms,and finally establishes a warning algorithm based on the combination of CCP(Car’s Current Position)and TLC(Time to Lane Crossing).The results of simulation experiments show that the algorithm can cope withdifferent types of lane departure events at different speeds.In Chapter 5,this thesis builds a single-preview-point driver model based on the magic formula tire model and the two-degree-of-freedom vehicle dynamics model,and then designs a lane keeping lateral control algorithm based on the optimal yaw rate under this driver model.In order to verify the feasibility of this algorithm,this thesis has conducted some path tracking experiments and lane keeping experiments on it under CarSim/Simulink.The experimental results show that the lateral control algorithm of this thesis has high accuracy for lane tracking,and it can control vehicles to run back to near the center line of lanes quickly and effectively when they deviate from lanes.