交通运输研究2024，Vol.10Issue(5)：78-90,13.DOI:10.16503/j.cnki.2095-9931.2024.05.007
高速公路车道级视频检测区自动设定方法
An Automatic Configuration Method for Video-Based Lane-Level Detection Zones on Expressways
摘要
Abstract
The paper proposed an automatic configuration method for lane-level detection zones on ex-pressways based on video,addressing the issue of the inability of Pan-Tilt-Zoom cameras to automatically set up video detection zones,which affected the accuracy of traffic event recognition.Firstly,by ana-lyzing the features of U-Net and MobileNet series models,combined with structures such as depthwise separable convolution and inverted residuals,a lightweight and efficient R-Net series model was de-signed specifically for semantic segmentation of lane lines and drivable areas.On this basis,according to the characteristics of expressways'specific scene recognition tasks,an algorithm for lane lines and detection zones labeling based on connected component analysis was proposed,which achieved auto-matic configuration of lane-level detection zones.At the same time,in order to improve the accuracy of the labeling algorithm,two preprocessing methods,threshold processing and frame stacking,were introduced for the first time.Then,a quadratic equation was used to fit the lane labeling results,achiev-ing complete and smooth segmentation of the lane lines.The experimental results showed that the per-formance metric MIoU of the R-Net series models was close to traditional models such as SegNet and U-Net,but significantly reduced the number of model parameters and inner product operations.Among them,the segmentation performance metric MIoU of the R-NetV2 model reached 90.6%,which was only 0.4%lower than U-Net,but its model parameter count was reduced by 38.7%and inner product operation was reduced by 62.5%.Labeling the preprocessed semantic segmentation results resulted in an increase in lane labeling accuracy from 80.47%to 95.58%compared to traditional methods.关键词
交通事件识别/视频检测/车道级检测区/轻量化/语义分割/连通域/车道线Key words
traffic event recognition/video detection/lane-level detection zone/lightweight/se-mantic segmentation/connected component/lane line分类
交通工程引用本文复制引用
莫宇蓉,吴烈阳,彭锦辉,刘圣卿,唐先亮,黎川,符锌砂..高速公路车道级视频检测区自动设定方法[J].交通运输研究,2024,10(5):78-90,13.基金项目
国家自然科学基金项目(51978283) （51978283）
江西省交通运输厅科技项目(2022X0037) （2022X0037）
