Research And Application Of Lane Detection Method Based On Attention Mechanism

With the development of economy and the progress of society,urban traffic has been continuously improved,and auto-driving and assisted driving systems of automobiles have emerged as the times require.Indispensable to such systems is lane detection technology.At present,there are many traditional methods for lane detection by manually selecting features,but such methods can only handle simple driving scenarios,and the detection speed is slow.In order to improve the accuracy of the model and the speed of detection,many researchers have introduced deep learning methods.In view of the difficulties and challenges of the current lane detection problem,the research work carried out in this paper is as follows:Aiming at complex road scenes,this paper improves the traditional spatial attention module,and introduces an efficient channel attention module at the same time,which is connected to form a fusion attention module.After adding this module to the network,the deep feature information of the image can be extracted to detect lanes more accurately.The network structure of this article includes the main branch and auxiliary branch.The main branch is responsible for extracting image features and obtaining high-level features under the action of the fusion attention module;the auxiliary branch is responsible for supervising training,connecting the extracted multi-layer features,and then convolution Layer to get the final test result.In this thesis,a lane detection system is designed.Through the investigation of the existing system,demand analysis and overall design are carried out.In the design stage,the entire system is divided into two parts,and the traditional method and deep learning method of lane detection are carried out Design and implementation of the process.Traditional methods include image preprocessing operations and lane fitting.Deep learning methods include network model building,training,use,and the visualization of detection results.In the testing phase,the system functions,model performance and system requirements for the operating environment were tested,and the expected standards were reached.For the lane detection model proposed in the article,ablation experiments and comparison experiments were performed on the CULane public data set,and certain conclusions were drawn through the analysis of the experimental results.Experiments show that the lane detection model proposed in this thesis not only performs well in normal road scenes,but also can adapt to some complex road scenes,which verifies the effectiveness and robustness of the model in this thesis.