AI-Based Smart Traffic Monitoring and Violation Detection System: Design and Experimental Evaluation of a Prototype for Mogadishu, Somalia

Abstract

The rapid urbanization and increasing traffic density in Mogadishu necessitate advanced technological interventions to enhance urban mobility and road safety. This study presents the design and implementation of an AI-Based Smart Traffic Monitoring and Violation Detection System, a prototype engineered to automate the identification of traffic infractions, such as driving in the wrong direction and illegal parking. The system architecture integrates a Pythonbased computer vision pipeline that utilizes the YOLOv8 object detection model for vehicle localization and the EasyOCR engine for automated license plate recognition. Data captured from the simulation environment were processed in real time and stored in a MySQL relational database that supports a PHP-based administrative dashboard designed for law enforcement oversight. This study employed an experimental multi-methodological approach, gathering primary data through experimental simulations within a controlled laboratory environment using miniature road models and toy vehicles. The performance analysis focused on the detection accuracy, OCR recognition rates, and system latency to validate the efficacy of the prototype. The system achieved a vehicle detection accuracy of 95.0%, correctly flagged 92.0% of wrong-direction driving incidents and 90.0% of illegal parking violations, and attained a license-plate OCR recognition rate of 88.5% (rising to over 93.0% for stationary vehicles), with an average end-to-end system latency of 450 milliseconds from violation capture to dashboard alert. The results indicate that the proposed system can help bridge the gap between manual traffic monitoring and automated enforcement, providing a cost-effective framework for smart city infrastructure in developing urban contexts, and indicate that open-source computer vision and OCR tools can deliver near-real-time violation detection even in resourceconstrained environments.