摘要
为提高复杂场景下车道线检测的精度,提出一种基于多尺寸分解卷积的车道线检测模型。首先采用ResNet34网络作为编码器提取图像局部特征;然后基于分解卷积的原理设计多尺寸分解卷积残差模块,进一步提取多尺度的车道线特征;最后设计车道线预测分支和基于双线性插值的解码器来分别对车道线进行置信度预测和像素级分类输出。在CULane数据集上对模型进行验证,平均调和均值F1达到75.3%,并在实际道路上进行了测试,结果表明,提出的模型在复杂场景下可以有效检测车道线。
In order to improve the accuracy of lane detection in complex scenes,this paper proposed a lane detection model based on multi-size factorized convolution.Firstly,ResNet34 network was used as an encoder to extract local features of the image,and then a multi-size factorized convolution residual module was designed based on the principle of factorized convolution to further extract multi-scale lane features,and finally the lane prediction branch and the bilinear interpolation-based decoder were designed to perform confidence prediction and pixel-level classification output for lane lines respectively.The model was validated on the CULane dataset and the average F1 index reached 75.3%,and it was tested on real roads.The results show that the proposed model can effectively detect lane lines in complex scenarios.
作者
李守彪
武志斐
Li Shoubiao;Wu Zhifei(Taiyuan University of Technology,Taiyuan 030024)
出处
《汽车技术》
CSCD
北大核心
2022年第8期32-37,共6页
Automobile Technology
基金
山西省回国留学人员科研资助项目(2021-050)。
关键词
分解卷积
车道线检测
语义分割
自动驾驶
Factorized convolution
Lane detection
Semantic segmentation
Automated driving