摘要
针对夜间场景下车道线检测存在的曝光度低、车道线模糊导致查全率低的问题,为提高基于深度学习的车道线检测方法在夜间低曝光场景下的车道线检测的查全率,提出了一种深度图像增强网络和车道线检测网络(LaneNet)相结合的车道线检测方法。首先,使用一个基于生成对抗网络(GAN)的图像增强网络对待检测图像进行增强,提高图像对比度同时增强其梯度;然后,使用一个基于编码器解码器架构的车道线检测网络LaneNet进行车道线检测并进行实例分割。实验结果表明,该方法在白天场景下表现与LaneNet相当,但在夜晚场景下的车道线正确检出数比LaneNet提高了27%,这表明该方法有效提升了了原车道线检测网络LaneNet在夜晚场景下的查全率。
Lane detection under night condition has a problem of low recall due to low visibility.In order to improve the recall of the deep learning based lane detection method under night condition,a lane detection method based on Deep Neural Network(DNN)which combined image enhancement technique and lane detection Network(LaneNet)together was proposed.Firstly,an image enhancement network based on generative network(GAN)was used to improve the gradient and contrast of the image,Then,the lane detection network LaneNet was used to detect the lane.The experimental results show that the proposed method has equal performance compared to LaneNet during daytime,but its accuracy under night condition was improved by 27%,which indicated that the recall of the origin lane detection network LaneNet has been improved dramatically.
作者
宋扬
李竹
SONG Yang;LI Zhu(School of Electronic&Information,Hangzhou Dianzi University,Hangzhou Zhejiang 310018,China)
出处
《计算机应用》
CSCD
北大核心
2019年第S02期103-106,共4页
journal of Computer Applications
关键词
图像增强
夜间
车道线检测
深度学习
实例分割
image enhancement
night condition
lane detection
deep learning
instance segmentation
