Research On Lane Detection And Deviation Warning Algorithm Based On Deep Learning

Lane detection and departure warning are important components of advanced driver assistance systems,and are crucial for improving driving safety.Aiming at the problems of insufficient accuracy and robustness of lane line detection algorithm in complex pavement environment,The real-time and accuracy of the departure of early warning needs to be improved,This thesis focuses on research related to lane detection and departure warning based on deep learning technology,The specific details are as follows:(1)To address the low detection rate and poor robustness of traditional lane detection algorithms under complex driving conditions,a deep learning approach is employed to detect lane lines.Specifically,the LaneNet network is used as the basis,with cascaded atrous spatial pyramid pooling and mixed attention modules in the encoding part,which can enhance the network’s ability to extract and fuse lane line features of different scales.And to improve the expression of lane line detail information.The improved LaneNet network achieves a recognition rate of 97.1% on the Tusimple dataset.(2)To promptly determine vehicle deviation and shorten the deviation warning time,a lane departure prediction network based on DAM＿LSTM is constructed.This network combines the LSTM with spatial and temporal dual attention mechanisms,and predicts the yaw angle and lateral distance of the driving vehicle in future moments.Experimental testing indicates that this algorithm achieves an accuracy rate of 94.95% and 95.21% in predicting the yaw angle and lateral distance,respectively.(3)A lane departure warning system was designed and implemented by combining a lane detection model with a lane departure prediction model,and adopting a dual alarm strategy for lane departure.The feasibility of the proposed algorithm was demonstrated by conducting system functional tested on both urban and highway structured roads.The thesis provides technical support for the detection and deviation warning of the visual lane line in complex scenarios.