Research On Real-time Lane Detection Algorithm Based On Convolutional Neural Network

The lane line detection occupies a large proportion in the environment perception of self-driving cars.The method of lane line detection based on convolutional neural network is an important research direction of the environment perception of self-driving cars.This paper designs a convolutional neural network to complete the detection and recognition of lane lines in car driving scenarios,and compares it with the traditional lane line detection methods constructed in this paper.The experiment shows that the convolutional neural network lane detection methods constructed in this paper has better accuracy and is more robust.The specific research contents are as follows:First,the data set used in the training process of the convolutional neural network model in this paper is established.Explain the data source and data annotation tool of the data set created in this paper,propose the annotation method of the road image in this paper,and the method of converting the file generated after the annotation to meet the data requirements of the data set in this paper,and finally convert to Tfrecord format through batch storage.Then build the convolutional neural network in this paper.First,configure the software environment for constructing the convolutional neural network,and then design the entire convolutional neural network using the encoder-decoder network structure.The encoder network of the convolutional neural network designed in this paper is a feature extraction network.Resnet34 is selected as the feature extraction network of the convolutional neural network in this paper.The decoder network is a two-branch network.The two-category branch completes the segmentation of the lane line area and the background area in the road image.The pixel embedding branch initializes a pixel embedding vector for each pixel and performs distance metric learning.Finally,the Meanshift clustering algorithm is used to cluster the output results of the convolutional neural network to segment the pixels of different lane lines and label them accordingly.Finally,the lane line detection results are obtained by curve fitting and regression.And design and implement traditional lane line detection algorithm.Carry out image preprocessing on the input road image by image preprocessing methods such as area of interest division,graying,and image filtering,and then use the automatic threshold method to divide the lane lines,and curve fit the divided lane lines.Finally,the traditional lane line detection algorithm constructed in this paper is compared with the convolutional neural network lane line detection algorithm designed in this paper,which shows that the convolutional neural network lane line detection algorithm designed in this paper has better accuracy and is more robust.