Research On Lane Detection And Lane Departure Warning System Based On Deep Learning

The rapid spread of automobiles has brought about increasingly serious traffic problems,and traffic safety has become one of the most concerned social issues.Among the many factors that cause traffic accidents,human factors are the most important ones.As an important part of the advanced driver assistance systems,the lane departure warning system can predict in advance whether the vehicle is about to departure from the lane,and issue an alarm signal to remind the driver to correct the driving state of the vehicle,which can effectively avoid traffic accidents caused by unintentional lane departure.In order to establish the lane departure warning system,this thesis makes an in-depth study on lane detection algorithm,lane tracking algorithm and lane departure warning model.A lane detection model based on deep learning is established,and lane line tracking and lane departure warning are implemented on this basis.The main research contents and results of the thesis are as follows:?.A lane detection model based on multi-task convolutional neural network is established.The model adopts the dilated convolution pyramid network.According to the "anchor grid" idea,the lane detection is transformed into a multi-task model of lane line type classification and key point coordinate regression to realize end-to-end lane detection.At the same time,a multi-scene lane detection dataset and multiple evaluation indicators are designed to test and analyze the model.?.Through the calibration of the internal and external parameters of the on-board camera and the inverse perspective transformation process,the lane lines in the perspective image are converted to the inverse perspective image.Then,a quadratic polynomial fit is applied to the set of lane marking points to simplify the lane line model.?.The Kalman filter is used to track the lane line fitting curve under the inverse perspective image,so that the lane lines between consecutive frames are consistent.It can reduce the interference caused by the lane line false detection or missed detection of a few frames.?.A lane departure warning model based on dynamic departure threshold is proposed.The vehicle lateral departure threshold is dynamically adjusted according to the vehicle yaw angle.The vehicle departure state is comprehensively determined from the vehicle lateral departure distance and the departure threshold.Combined with the departure frame counter,it is finally determined whether to issue a departure alarm.?.A complete lane departure warning system is built and the full functionality of the system is implemented in the Ubuntu system.Finally,the system is tested and analyzed under the simulated driving environment established by PreScan simulation software and real vehicle environment.The test results show that the lane departure warning system has good accuracy and robustness,and can meet the real-time requirements of the system.