基于非对称卷积的多车道线检测方法

Multi-Lane detection method based on asymmetric convolution

针对车道检测的准确性和实时性之间不平衡的问题,构建一个基于Lanenet算法和图像增强技术的多车道线检测网络,旨在更全面地利用图像中的特征信息,提高检测精度和速度。使用多尺度Retinex算法对输入图像进行色彩增强、降噪等;设计采用一种双边多尺度融合网络实现浅层特征与深层特征之间的信息交互,获取上下文语义。提出一个新的非对称卷积金字塔模块,将非对称卷积融合到不同扩张率的空洞卷积层中,提高网络的特征提取能力,减少计算量。实验结果表明,该方法与现有的深度学习算法相比,能够在遮挡和阴影条件下更有效地检测车道线,具有更高的精度,更低的误检率和漏检率。

Aiming at addressing the imbalance between the accuracy and real-time was of lane detection,a multi-lane detection network based on Lanenet and image enhancement technology was constructed to make use of feature information in the image and improve the detection accuracy and speed.Multi-scale Retinex algorithm was used to enhance the color of the input image and reduce noise.A bilateral multi-scale fusion network(BMFNet)was designed to realize the information interaction between shallow features and deep features and capture the context semantics.A new asymmetric convolution pyramid module(ACP)was used to fuse asymmetric convolution into atrous convolution layers with different dilated rates,so as to improve the feature extraction ability of the network and reduce the amount of computation.Experimental results show that compared with the existing deep learning algorithms,the proposed method can effectively detect lane under occlusion and shadow conditions,and has higher accuracy,lower false positive(FP)and false negative(FN).