Research On Lane Line Recognition Technology

With the development of economy and science and technology,people's living standards are improving day by day,urban traffic is constantly improving,and the number of cars has increased significantly.At the same time,various types of traffic accidents caused by cars are also increasing.In order to improve the safety of cars and reduce the occurrence of traffic accidents,driverless technology came into being.As a basic application of artificial intelligence technology,driverless technology is gaining more and more attention from scientific researchers and technology companies.The accuracy of lane line recognition directly determines the success or failure of driverless technology,which is the core part of driverless technology.On a road with clear weather,sufficient and uniform light,clear lane lines,and good road conditions,identifying lane lines may not be difficult,but in the real world,such road conditions are not often seen.How to identify lane lines under severe conditions has always been a difficult point to be overcome in the field of unmanned driving,and it is also a very hot research topic in this field.Based on this research topic,this paper has done the following work:(1)A series of simple lane line recognition algorithms are used to identify lane lines in simple scenarios.Under the condition that the recognition environment is excellent and the lane lines are clear and unobstructed,the lane lines are identified by related algorithms according to the "shape" and "color" of the lane lines.This method is fast in recognition and does not have complicated calculations.However,the robustness is not strong,and the requirements for the identified scene are more stringent.(2)Use Hough transform to identify straight lane lanes.Huff transform is a classic edge recognition algorithm that can be used to detect straight lines and analyzable curves.This method is fast and has high accuracy,but it still has certain requirements for recognizable scenes,especially for scenes with many curves such as urban viaducts,the recognition effect is poor,so the robustness is average.(3)Aiming at the shortcomings of Hough transform,the existing lane line detection method based on Hough transform is improved.It can also have a good recognition effect for curve parts that cannot be easily fitted with analytical curves.In addition,the Huff transform is combined with the first method described above.By performing a certain preprocessing on the original picture,the improved Huff transform not only has a satisfactory recognition speed,but also recognizes scenes more complex and hasstronger robustness.(4)The deep learning method is used to abandon the intuitive "shape" and "color" features of the lane lines and extract the high-dimensional features of the lane lines to achieve better recognition results.By building a LaneNet recognition framework and using the TuSimple dataset for training,lane lines can be detected under more complex road conditions.Compared with Hough transform,the accuracy of recognition,and acceptable scene complexity have all taken a qualitative leap.