摘要
近年来,车道路面的识别与语义分割已成为无人驾驶领域的研究热点。本文针对现有的CNN算法在解决前述问题时存在的不足,提出了一种带有空间信息的卷积神经网络车道路面语义分割模型。在传统的卷积神经网络中,任意层负责接收上层的数据,再作卷积并加激活传给下一层。这个过程是顺序执行的,但是对于车道路面识别问题而言,采用传统的卷积神经网络无法很好地提取前方路面的空间信息。本文提出的模型将特征图的每一行或每一列作为卷积层的输入,这使得空间信息能够在同层的神经元上传播。该模型在车道路面语义分割任务中的准确率,明显高于FCN、U-Net等传统语义分割模型。
In recent years, road surface recognition and semantic segmentation have become research hotspots in the field of unmanned driving. Aiming at the shortcomings of the existing CNN algorithms in solving the aforementioned problems, this paper proposes a convolutional neural network road surface semantic segmentation model with spatial information. In the traditional convolutional neural network, any layer is responsible for receiving the data of the upper layer, and then convolving and adding activation to the next layer. This process is performed sequentially, but for the road surface recognition problem, the traditional convolutional neural network cannot extract the spatial information of the road ahead. The model proposed in this paper takes each row or each column of the feature map as the input of the convolutional layer, which enables the spatial information to propagate on neurons in the same layer. The accuracy of this model in the task of semantic segmentation of road surface is significantly higher than that of traditional semantic segmentation models such as FCN and U-Net.
作者
敖凌文
刘婷
AO Lingwen;LIU Ting(Hunan College of Information,Changsha Hunan 410200,China)
出处
《信息与电脑》
2021年第2期43-45,共3页
Information & Computer
关键词
语义分割
识别
卷积神经网络
semantic segmentation
recognition
convolutional neural network
