Unauthorized vehicles in dedicated bus lanes disrupt public transit efficiency and safety. To overcome the limitations of manual monitoring, this study proposes a real-time computer vision pipeline to detect TransJakarta lane violations. The system utilizes the lightweight YOLOv11n model for accurate vehicle detection and evaluates two Multi-Object Tracking (MOT) algorithms: ByteTrack and SORT. YOLOv11n delivered strong detection performance, achieving an F1-Score of 0.8078, an mAP@0.5 of 0.8630, and 55.8 FPS. In the tracking phase, ByteTrack significantly outperformed SORT, tracking 2,535 vehicles and detecting 372 violations at a rapid 157.6 FPS, compared to SORT’s 1,751 vehicles and 309 violations at 44.6 FPS. ByteTrack’s efficiency is driven by its ability to associate both high- and low-confidence detection boxes, which minimizes track fragmentation. Conversely, SORT discards low-confidence detections, resulting in broken trajectories and more conservative outputs. Ultimately, integrating YOLOv11n with ByteTrack provides a robust and computationally efficient solution tailored for real-time automated traffic enforcement.
Copyrights © 2027