Abstract
The increasing problem of urban traffic congestion remains a major challenge for rapidly growing cities across the world, especially in developing countries where infrastructure development struggles to keep up with the pace of rising number of vehicles. In light of the limitations of manual traffic monitoring and the high costs of sensor-based systems, this paper introduces a deep learning-based approach for real-time vehicle detection and tracking, which is tailored for environments where resources are limited. This study utilises YOLOv8 which is an object detection framework combined with the DeepSORT tracking algorithm to capture, classify, and count vehicles under real-world conditions. The study used a custom video dataset which was recorded along the Sokoto–Samaru expressway in Zaria, Nigeria, encompassing seven vehicle classes used for training and evaluation. Evaluation metrics included per-class precision, recall, F1-score, and mean Average Precision (mAP) at IoU thresholds of 0.5 and 0.5:0.95. Results show strong performance for motorcycles (F1 = 0.973), tricycles (F1 = 1.000), and cars (F1 = 0.833), with an overall mAP@0.5 of 65.0% and mAP@0.5:0.95 of 55.3%. However, result also revealed underperformance for vans, buses, and pickups, which is attributed to dataset imbalance and visual similarity between some classes of vehicles. The proposed system's competitiveness in real-time ITS applications is demonstrated through comparison with state of-the-art methods, revealing potential for enhancement through dataset augmentation, small-object detection modules, and multimodal fusion. These findings position the proposed system as a viable baseline cost-effective, scalable traffic monitoring solutions in developing urban contexts.

National Library of Nigeria
Association of Nigerian Authors
Nigerian Library Association
EagleScan
Crossref