@Article{cmc.2026.086550,
AUTHOR = {Mohammed Alonazi, Muhammad Adeel Ahmed Tahir, Adnan Ahmed Rafique, Maha Abdelhaq, Raed Alsaqour, Ahmad Jalal, Jeongmin Park},
TITLE = {Intelligent Urban Transportation over Complex Vehicle Networks with YOLOv8 for Traffic Flow Monitoring},
JOURNAL = {Computers, Materials \& Continua},
VOLUME = {},
YEAR = {},
NUMBER = {},
PAGES = {{pages}},
URL = {http://www.techscience.com/cmc/online/detail/27980},
ISSN = {1546-2226},
ABSTRACT = {Accurate vehicle detection, tracking, and counting are fundamental components of Intelligent Transportation Systems (ITS) and urban traffic surveillance. However, real-world deployment remains challenging due to domain shifts, illumination variations, occlusions, dense traffic conditions, and heterogeneous data distributions. Existing studies often address detection, tracking, and counting as independent tasks, resulting in limited cross-domain generalization and inconsistent performance in complex traffic environments. To overcome these limitations, this paper proposes a unified cross-domain framework that jointly integrates vehicle detection, tracking, and lane-aware counting within a single intelligent traffic analytics pipeline. The proposed framework begins with image enhancement using Contrast Limited Adaptive Histogram Equalization (CLAHE) and Gamma Correction to improve visual quality under varying illumination conditions. A Domain-Adversarial Neural Network (DANN) equipped with a Gradient Reversal Layer is employed to learn domain-invariant feature representations across heterogeneous datasets. Vehicle localization is performed using a dual-stage detection strategy in which YOLOv8 generates high-recall object proposals and RetinaNet refines candidate detections to reduce false positives and improve localization accuracy. For multi-object tracking, ByteTrack is enhanced with appearance-based Re-Identification embeddings and Kalman filtering to improve identity preservation and trajectory consistency under occlusions and crowded traffic conditions. Furthermore, a lane-aware multi-zone counting mechanism is introduced to replace conventional line-crossing approaches by enforcing trajectory-consistent spatial transitions across lane-aligned regions. Experimental evaluation on the Cityscapes and KITTI datasets demonstrates significant improvements in detection accuracy, tracking robustness, identity preservation, cross-domain generalization, and vehicle counting performance. Ablation studies further validate the contribution of each module to the overall framework. The proposed system provides an effective solution for real-time urban traffic monitoring and intelligent transportation analytics across heterogeneous environments.},
DOI = {10.32604/cmc.2026.086550}
}