基于编码-解码网络的车道线检测算法

Encoding-decoding model based lane detection algorithm

为解决在车道线磨损、被遮挡以及光照变化等复杂场景中车道线检测精度较低的问题,提出了一种基于编码-解码网络的车道线检测算法。首先,对Resnet18网络进行改进和优化,组成编码网络;然后,结合ASPP模块和金字塔注意力机制组成解码网络,对图像进行像素级的语义分割,识别并区分车道线;最后基于自适应拟合算法拟合车道线。在Tusimple公开数据集上进行训练和测试,结果表明,该算法的准确率、检测速率、误检率和漏检率分别为:96.45%、35帧/秒、2.59%、1.41%,在复杂场景下的检测精度较高,鲁棒性较强。

In order to solve the problem of low accuracy of lane detection in complex scenes such as lane wear,occlusion and light change,a lane detection algorithm based on encoding-decoding network is proposed.Firstly,the Resnet18 network is improved and optimized to form a coding network.Secondly,the ASPP module and pyramid attention mechanism are combined to form a decoding network to segment the image at pixel level,which is used to recognize and distinguish lane lines.Finally,the lane lines are fitted by adaptive fitting algorithm.The proposed method is trained and tested on Tusimple public data set.The results show that the accuracy,detection rate,false detection rate and missed detection rate of the proposed algorithm are 96.45%,35 frames/s,2.59%and 1.41%,respectively.The proposed method has high detection accuracy and strong robustness in complex scenarios.