Research On Lane Detection Method Based On Convolutional Neural Network

The traditional lane line detection algorithm is susceptible to the environment,and has some problems such as low detection accuracy and slow detection speed.Deep learning has gradually become the mainstream method of lane line detection.Currently,the detection effect of some deep learning-based lane line detection algorithms on scenarios such as cloudy and rainy days,night and vehicle occlusion needs to be improved.In order to improve the reasoning speed and detection accuracy of the model and realize lane detection in various complex scenes,a lane detection algorithm LC-LaneNet was proposed by improving LaneNet algorithm and integrating attention module and multi-scale features.The main research contents are as follows:(1)Aiming at the lack of Tusimple night lane scene and the lack of CULane night lane scene,an urban night lane line dataset(CSData)is constructed.Firstly,a near-infrared camera was used to collect nighttime lane images in urban areas.Then the image was screened and labelme software was used for annotation processing according to Tusimple annotation format.Finally,the image processing algorithm is used to enhance and expand the day and night data sets to prevent the over-fitting phenomenon in the training process.(2)In order to improve the accuracy of LaneNet algorithm and give consideration to real-time performance,the overall scheme design is carried out in this paper,and a lane line detection algorithm LC-LANenet which integrates attention and multi-scale features is proposed.Firstly,the compression and convolution module reconstruction of LaneNet backbone network ENet are carried out to improve the ability of multi-scale feature extraction and network reasoning speed.Secondly,the improved spatial attention mechanism and channel attention mechanism are added to ENet network to enhance the connection between lane line location information,improve the situation that lane lines are not obvious or blocked,and improve the accuracy of network detection.Finally,Concat operation is used to connect shallow features and deep features of the improved ENet network to enhance the semantic information of the network context.(3)Experiment and analyze the LC-Lanenet algorithm proposed in this paper.The experimental results show that the detection accuracy of LC-Lanenet algorithm is 97.04%,and the detection speed is 66.7fps.Compared with LaneNet algorithm,the detection accuracy is improved by0.66%,and the detection speed is improved by 14.1fps.The visualization results of LC-Lanenet algorithm on three data sets show that the algorithm can adapt to the detection of most complex lane line scenes.