Work

A Motorized-Vehicle-Lane Region Segmentation Method for Detecting Illegal E-Bike Riding

Year: 2026
Type: article
Abstract: Motorized-vehicle-lane region segmentation under fixed-camera views is challenging due to perspective-induced scale variation, vehicle occlusion, and road texture interference. These factors can cause missing regions, segmentation discontinuities, and boundary localization errors, increasing the risk of the missed or incorrect detection of illegal e-bike riding. To address these issues, this study proposes an improved YOLO11n-seg-based method for motorized-vehicle-lane region segmentation. A Perspective-aware Directional Continuity Module (PDCM) is introduced at the high-resolution P3 feature level. It combines multi-scale and directional features to enhance road-feature representation in distant and occluded regions, reducing feature degradation and segmentation discontinuities. In addition, a region–boundary joint segmentation loss combines binary cross-entropy (BCE), Dice, and boundary Dice losses. It jointly constrains the region overlap and boundary structure to reduce boundary localization errors caused by road shadows, lane markings, and adjacent pavement textures. The experimental results show that the improved model achieves a recall of 79.691% and an [email protected]:0.95 of 59.156%, with an improvement of 3.002 percentage points in [email protected]:0.95 over the YOLO11n-seg baseline. The segmentation results are integrated with ground-contact-point-based spatial determination, ByteTrack trajectory association, and temporal-window-based occupancy statistics to detect illegal e-bike riding. Continuous surveillance video tests demonstrate that the extracted lane regions provide a stable spatial reference for violation detection.
Source: Electronics
Language: English
Open Access status: gold
Estimated APC paid: $2,896
OpenAlex ID: W7235239054
FWCI:
Cites: 31
Cited by:
Related to: 10