Lane Detection Based On Deep Learning

With the development of economy and society and the continuous improvement of people’s living standard,the car ownership keeps rising year by year.This not only causes serious environment pollution and traffic congestion problems,but also increases the number of traffic accidents year by year.As an important part of self-driving,lane detection is the focus of environment perception.Traditional lane detection algorithms based on vision require high experience and is sensitive to illumination,occlusion and interference of other external environment factors,resulting in weak robustness and adaptability of the algorithm.Against the disadvantages of the traditional lane detection algorithm,we combined deep learning with the post-progressing algorithms for lane detection in this paper.(1)Collect the pictures in different road conditions,different weather conditions,different traffic conditions and different lighting conditions to make training set,verification set and test set,combined with the image processing technology and annotation technology to mark and preprocess the collected images.(2)Framework of deep learning is used to build the convolutional neural network,the preprocessed data is input into the network for training.(3)Make use of the post-progressing algorithms to the segmentation result of the model with the peak value point extraction,clustering algorithm and make regression to the original lane.The validation set is used to evaluate the effect of the model,optimize and adjust the parameters of the model according to the actual and get the final model finally.With examination,the studied lane detection algorithm is more robust than the traditional lane extraction algorithm based on a single feature and adapted to a wide range of scenarios.