IMPLEMENTATION OF OBJECT DETECTION SYSTEM (VEHICLE MONITORING IN HEAVY URBAN TRAFFIC IN NIGERIA)

This study develops and evaluates a YOLOv8-based object detection system to enhance intelligent traffic monitoring in Nigeria, addressing challenges such as occlusions, varying lighting conditions, and limited computational resources. With over 10,000 annual road fatalities, Nigeria’s diverse traffic dynamics necessitate automated solutions beyond traditional manual surveillance. The research aims to achieve high accuracy and near-real-time performance, surpassing Zhang et al.'s benchmark mAP50 of 0.91. A curated dataset of over 5,000 images, annotated with four vehicle classes (car, truck, bus, motorcycle), was prepared using Label Studio and augmented with techniques like mosaic blending and HSV shifts. Implemented on a NVIDIA RTX 3060 GPU via Google Colab with Python 3.12 and PyTorch, the model was trained for 40 epochs using SGD optimization. Results demonstrate a mAP50 of 0.96, precision of 0.94, recall of 0.86, and an inference speed of 85 FPS, outperforming the benchmark. Despite a mAP50-95 of 0.72 indicating room for improvement with occluded vehicles, the system proves robust for real-time deployment. The study concludes with recommendations for dataset expansion and multi-modal integration, offering a scalable solution for safer and more efficient traffic management in resource-constrained settings.