Research On Lane Departure Detection System Based On Deep Learning

With the development of science and technology and economy,the improvement of road construction,the increasing amount of civil vehicles,the problem of traffic jam and traffic safety is getting more serious.Road safety has always been a hot topic.Topics related to Autopilot,Unmanned Driving,Advanced Driver Assistance System,etc,have always been hot parts in the field of transportation.According to the data statistics in recent years,most traffic accidents are not due to the quality of vehicles,it is because of fatigue driving and misoperation of drivers.So,if vehicles can provide active intervention mechanism,it can avoid many traffic accidents by giving warnings.Traffic accidents caused by lane changing account for about 40% of the total traffic accidents.Obviously Lane Departure Warning System is very significant.Lane departure warning system is one of the main research contents of Advanced Driver Assistance System.The detection technology of lane is the key of Lane Departure Warning System.The main contents of this thesis are lane detection technology,lane tracking technology and deviation judging technology and we combine these three parts into Lane Departure Detection System which not contains warning function.In terms of lane detection technology,we firstly used traditional visual method based on Hough transforming.In this method,the images are divided into a region of interest(ROI)and the rest,then we deal ROI with grayscale,Gaussian blur,binarization,etc.,and then we use.Canny edge detection algorithm and Hough transforming to detect lane lines.The accuracy of the experimental result is about 83%.However,the traditional visual method is prone to fail to detect lane lines when the lane lines are blurred,damaged,or blocked.And it cannot detect curves.In order to solve the above problems,this thesis introduces a deep learning method and uses a semantic segmentation network model to locate the pixel positions of lane lines more accurately,thereby we can achieve the purpose of identifying curves and improving the detection effect of lane lines that are blurred,damaged,or blocked.In this thesis,a lane segmentation network model based on a convolutional network is introduced.The pyramid pooling module,spatial attention module,self-attention module,and spatial convolution module are introduced to enhance the spatial context information of the long-distance structure of lane lines.And on this basis,the Kalman filter is used to track the lane lines,so that the lane lines in consecutive frames can maintain continuity,and the final detection effect of the system is not affected by missed or mis-detected lane lines.Beside,we set a threshold,and the relationship between the special distance and the threshold determines whether the vehicle deviates.Finally,a lane departure detection system is built,and the system is tested and analyzed in a computer environment.The results show that the lane departure detection system has good accuracy and robustness.