UFLD(ultra fast structure aware deep lane detection)是一种轻量化车道线检测模型,为提升模型的检测精度,对模型进行改进.引入CAM(channel attention mechanism)使模型能更关注携带重要车道线信息的特征通道和像素;为了感知车道线的细节信息,引入ASPP(atrous spatial pyramid pooling)扩大卷积过程的感受野,提高模型分割精度;搭建引入CAM和ASPP后的改进模型,并在改进的模型上进行实验.实验结果表明:在TuSimple数据集上以Res-Net18为主干网络的模型检测精度由95.81％提升至95.98％,以ResNet34为主干网络的模型检测精度由95.84％提升至96.12％;在CULane数据集上,无论是以ResNet18还是以ResNet34为主干网络模型,其平均精度均有不同程度的提高.
Research on lane line detection algorithm fusion CAM and ASPP
The UFLD ( Ultra Fast Structure Aware Deep Lane Detection ) is a lightweight lane line detection model.To improve the detection accuracy, this paper improves the model.First, the Channel Attention Mechanism ( CAM) is introduced to make the model more attentive to the feature channels and pixels that carry important information.Secondly, to perceive the detailed information of line lanes, the Atrous Spatial Pyramid Pooling ( ASPP) is introduced to expand the receptive field of convolution operation and improve the segmentation accuracy of model.Finally, an improved model is built after introducing CAM and ASPP, and experiments are conducted.Our experimental results show on the TuSimple dataset, the detection accuracy increases from 95.81％ to 95.98％ with ResNet18 as the backbone network, and rises from 95.84％ to 96.12％ with ResNet34 as the backbone.On the CULane dataset, whether the model uses ResNet18 or ResNet34 as the backbone network, its average accuracy improves with varied degrees.
国家科技期刊平台
NSTL
万方数据
