Research On Lane Detection Technology Based On Deep Learning

Lane detection plays an important role in lane departure warning,automatic driving and other fields.Traditional Lane detection performs road feature extraction based on specific scenes,so it is difficult to deal with complex and changeable road environments to a certain extent.With the rapid development of deep learning technology,more and more researchers are committed to deep learning-oriented lane detection technology,which has expanded the research process of lane detection.This paper uses deep learning technology to study the image-based and video-based lane detection technology,including:(1)Image-based Lane detection technology.Firstly,pre-process the public road line data set CULane through clustering analysis and manual annotation;secondly,use the cutting algorithm to cut the slender road line in the picture into several road line target blocks,thereby detecting the lane The problem is transformed into lane target block detection;further,the multi-scale deep learning network is used to extract the lane features of different scales,and the targeted anchor point scale is designed according to the relationship between the effective receptive field and the anchor point,and the anchor point matching is proposed.The strategy corrects the matching of the anchor point and the target frame to improve the recall rate of the lane detection;finally,the least square method is used to perform linear regression,and the center coordinates of the prediction frame are returned to a line to realize the line detection based on the target block method.(2)Video-based lane detection technology.First,an adaptive key frame technology is proposed on the basis of image-based line detection,and the key frames in the video sequence are determined by judging the sharpness of the image in the current video;secondly,the features between different video frames are extracted by using a recurrent neural network Information and fusion are carried out to strengthen the transmission of information between frames;finally,based on the attention mechanism,the key elements related to the lane are emphasized,so as to achieve more accurate lane detection.Experimental results show that the image-based and video-based lane detection techniques proposed in this paper have achieved breakthrough detection results,effectively improving the accuracy of line detection.