基于MFLD的车道线识别方法  

Lane Line Detection Method Based on MFLD

作　　者：段慧林 孙健力 张建臣 DUAN Huilin;SUN Jianli;ZHANG Jianchen(School of Information and Control Engineering,Jilin Instit ute of Chemical Technology,Jilin Jilin 132022,China;School of Computer Science and Information,Dezhou Un iversity,Dezhou Shandong 253000,China;Shandong Vocational College of Electronics,Jinan Shandong 250200,China)

机构地区：[1]吉林化工学院信息与控制工程学院,吉林吉林132022 [2]德州学院计算机与信息学院,山东德州253023 [3]山东电子职业技术学院,山东济南250200

出　　处：《德州学院学报》2024年第4期33-38,75,共7页Journal of Dezhou University

摘　　要：在无人驾驶领域中,车道线检测是一个至关重要且具挑战性的任务。传统的视觉车道线检测方法处理缓慢、操作复杂,需要人工干预,而基于深度学习的卷积神经网络方法能够有效地克服这些问题,为无人驾驶技术的进步提供了关键推动力。基于卷积神经网络进行车道线检测的方法有很多,近年来比较流行的方法中ultra fast lane detection(UFLD)算法在车道线检测方面取得较好的成果。文章以UFLD算法为基础,结合MobileNetV2和CBAM注意力机制,设计出了一种新的模型,命名为mobilenet-fast lane detection(MFLD),该模型在Tusimple检测准确度达到92.8%,相较于UFLD提高了2.3%,同时运行速度达到了129 fps,在CULane数据集检测精度在不同场景下相较于UFLD提升了2.5%、2.5%、2.9%、1.8%、1.0%、2.5%、1.3%、5.1%、6.4%。In the field of autonomous driving,lane detection is a crucial and challenging task.Traditional visual lane detection methods are slow,complex,and require manual intervention,whereas convolutional neural network(CNN)-based deep learning methods can effectively overcome these issues,providing a key impetus for the advancement of autonomous driving technology.There are many methods for lane detection based on convolutional neural networks,and among the popular methods in recent years,the Ultra Fast Lane Detection(UFLD)algorithm has achieved good results in lane detection.This paper designs a new model named MobileNet-Fast Lane Detection(MFLD)based on the UFLD algorithm,combined with MobileNetV2 and the Convolutional Block Attention Module(CBAM)attention mechanism.This model achieves a detection accuracy of 92.8%on the Tusimple dataset,improving by 2.3%compared to UFLD,while the operating speed reaches 129 FPS.On the CULane dataset,the detection accuracy improves by 2.5%,2.5%,2.9%,1.8%,1.0%,2.5%,1.3%,5.1%,and 6.4%in different scenarios compared to UFLD.

关 键 词：车道线检测 UFLD 注意力机制 深度学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]