Design Of Lane Line Perception System In Complex Environment Based On Monocular Vision

In recent years,with the development of social economy,science and technology,the traffic environment of the city has been continuously improved,and the number of cars has increased obviously.At the same time,traffic accidents are also rising year by year,which poses a great threat to the safety of people.Under this background,the research and development of intelligent safe driving technology is favored by more and more automobile enterprises and Internet enterprises,and a series of technology research with driverless background is paid more and more attention by academia and industry.In order to successfully realize the intelligent driving of a car on a structured road,one of the most important steps is that the system can detect lane information accurately and quickly from the current road image.However,these tasks can be very challenging due to the disturbance of many complicated conditions,such as changing light,shading of the environment,bad weather and so on.As a result,the robustness of the algorithm is reduced,and the required detection accuracy is not achieved in bad weather and complex environment.Aiming at the robustness of lane detection in complex environment at present,this paper researches and designs a system,which captures images by Basler Monocular camera(vehicle camera),and processes the images to realize lane detection in complex environmentFirstly,the system realizes image acquisition by Basler monocular camera,and then corrects image distortion and extracts the region of interest(ROI).Secondly,the extracted images of the region of interest(ROI)of lane lines are segmented by the lane line segmentation algorithm researched in this paper.Based on the traditional adaptive threshold algorithm for lane line segmentation,a method of lane line pixel segmentation based on depth learning image semantic segmentation is researched in this paper.The main method is based on VGG16-FCN binary segmentation network and pixel embedding layer(Embedding)distance metric learning network to realize lane line pixel segmentation.Then,we set up the lane line model of cubic B-spline curve and fit it by inverse perspective transformation and least square method to realize lane line detection.Based on the results of lane detection,the position and pose calculation of intelligent driving vehicle is realized.Finally,this paper designs multi-group test experiments including different data sets,different complex environments and different resolutions.The segmentation performance,accuracy performance and real-time performance of the system algorithm are tested respectively.The experimental results show that the proposed system and related algorithms are robust and can adapt to the disturbance of many complicated conditions,such as light variation,shadow shading,bad weather and so on.The accuracy of lane detection of the designed system can reach more than 89%,and the detection speed can reach about 13Hz.