Analysis of Lane Line Defect Detection Technology Based on Optimized DeepLab v3+
This paper describes a lane line defect detection method based on an improved DeepLab v3+semantic segmentation model.Replace the original Xception network with MobileNet v3 network to reduce the number of network parameters.At the same time,the dual attention mechanism CBAM is introduced after extracting the backbone features,and the lane defects are detected through the method of minimum rectangle bounding.Using a self-made dataset,train,validate,and test the improved DeepLab v3+based on Python.The results show that the improved DeepLab v3+improves MIoU and MPA by 1.6％and 1.3％respectively,with a single image segmentation time of 7.4ms,which is 16.9ms less than the original model.It can meet the real-time and accuracy requirements for lane line defect detection.
lane line defect detectionsemantic segmentationDeepLab v3+networkdual attention mechanism CBAM
NETL
NSTL
万方数据
