Road Line Marking Segmentation Using Multi Variants of YOLOv8 and YOLOv11 in CCTV Traffic Videos

Vehicle speed measurement plays a pivotal role in terms of detecting speed violations by the transport user. To measure vehicle speed in CCTV camera manner, a homography transformation network is used to transform video scene as a bird-eye view paradigm. However, an approach that can help initiate this algorithm only once when processing the first frame of a video is needed. This paper proposed segmentation of the road line marking to help produce homography transformation to measure vehicle speed in the CCTV video scene. YOLOv8 and YOLOv11 with small, medium, large, and extra large variants are used to segment road line marking from road line marking data set. The results of the segmentation metrics show that all variants of YOLOv8 beat all variants of YOLOv11 in the majority of the segmentation metrics. Then, YOLOv8 medium has the best mAP50 with value of 0.868 while YOLOv8 large has the best mAP50-95 with value of 0.468. The inference analysis result also shows that the large variant of YOLOv11 beat all variants of YOLOv8 and YOLOv11 in test images which are totally different from the dataset given.