Improved Lane Detection Algorithm Based on UFS Network
石磊 1范英 1苏伟伟 1闫浩1
扫码查看
点击上方二维码区域，可以放大扫码查看
作者信息
1. 太原科技大学 交通与物流学院,太原 030024
折叠
摘要
为了解决车辆行驶中面对各种复杂环境车道线检测算法精度不高的问题,提出一种基于改进的UFS网络检测算法(Ultra Fast Structure-aware Deep Lane Detection,UFS).首先,采用改进的Gamma校正对待检路面图像进行校正,降低光照、阴影等的影响,以提升夜间图像纹理特征.然后引入非局部神经网络模块(Non-Local Block),充分提取图像全局特征,以提高检测可靠性.最后对改进后的算法使用Tusimple、CULane数据集进行测试.结果表明:改进后的模型在物体遮挡、光照变化、阴影干扰等复杂场景下,提升了对复杂噪声与多元场景的处理能力,车道分割的准确率有所改善,具有较好的鲁棒性.
Abstract
In order to solve the problem of poor real-time and low precision of lane line detection algorithm in vari-ous complex environments,an improved UFOS network detection algorithm(Ultra Fast Structure-aware Deep Lane Detection,UFS)is proposed.First of all,the improved gamma correction is used to correct the inspection road image,reduce the influence of lighting,shadows,etc.,in order to enhance the texture characteristics of the night image.Then a non-local neural network module(Non-Local Block)is introduced to extract the global characteris-tics of the image in order to improve detection reliability.Finally,the improved algorithm is tested using TuSimple and CULane datasets.The results show that the improved model improves the processing power of complex noise and multiple scenes under complex scenes such as object masking,lighting change and shadow interference,and the accuracy of lane segmentation is improved,and it is more robust.
关键词
车道线检测/Gamma校正/UFS网络/非局部神经网络/CULane数据集
Key words
lane detection/Gamma correction/UFS model/non-local neural network/CULane dataset
维普
万方数据