融合CBAM注意力机制与可变形卷积的车道线检测  

Lane line detection incorporating CBAM mechanism and deformable convolutional network

作　　者：胡丹丹[1] 张忠婷 牛国臣[1,2] HU Dandan;ZHANG Zhongting;NIU Guochen(Robotics Institute,Civil Aviation University of China,Tianjin 300300,China;Key Laboratory of Smart Airport Theory and System,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学机器人研究所,天津300300 [2]中国民航大学民航智慧机场理论与系统重点实验室,天津300300

出　　处：《北京航空航天大学学报》2024年第7期2150-2160,共11页Journal of Beijing University of Aeronautics and Astronautics

基　　金：天津市科技计划(17ZXHLGX00120);中央高校基本科研业务费专项资金(3122022PY17,3122017003)。

摘　　要：为满足自动驾驶及汽车高级驾驶辅助系统(ADAS)对车道线检测准确性和实时性的要求,提出一种融合卷积块注意力机制(CBAM)与可变形卷积网络(DCN)的车道线检测方法CADCN。在特征提取模块中嵌入CBAM注意力机制,增强有用特征并抑制无用特征响应;引入可变形卷积替换常规卷积,用带偏移的采样学习车道线的几何形变,提高卷积核的建模能力;基于行锚分类思想,对行方向上的位置进行选择和分类分析,预测车道线的位置信息,提高车道线检测模型的实时性。在车道线公开数据集上对所提CADCN方法进行训练及验证,在满足实时性的情况下,CADCN方法在TuSimple数据集上准确率达到96.63%,在CULane数据集上综合评估指标F1平均值达到74.4%,验证了所提方法的有效性。In order to meet the accuracy and real-time requirements of autonomous driving and advanced driver assistance systems(ADAS) for lane line detection,a CADCN lane line detection method incorporating convolutional block attention module(CBAM) mechanism and deformable convolutional network(DCN) was proposed.Firstly,the CBAM mechanism was embedded in the feature extraction module to enhance the useful features and suppress the useless feature responses.Secondly,DCN was used to replace the conventional convolutional network,and the geometric deformation of lane lines was learned by sampling with offset to improve the modeling capability of the convolution kernel.Finally,based on the idea of row anchor classification,the location point along the row was selected and classified,so as to predict the lane line location information and thus improve the real-time performance of the lane line detection model.The CADCN model was trained and validated on the public lane line dataset.While ensuring real-time performance,the accuracy rate of the model on the TuSimple dataset reaches 96.63%,and the comprehensive evaluation index F1 on the CULane dataset reaches 74.4%,which verifies the effectiveness of the algorithm.

关 键 词：车道线检测 特征提取 注意力机制 可变形卷积网络 行锚分类 

分 类 号：U471.11[机械工程—车辆工程] TP391.4[交通运输工程—载运工具运用工程]