Computer Vision Analysis for Traffic Monitoring and Road Safety in Smart City Concept

Abstract

Rapid urban growth and rising traffic complexity require Smart City solutions that move beyond passive CCTV toward intelligent, real-time traffic management. This study examines how computer vision–based analytics contribute to road safety when integrated into an Intelligent Transportation System (ITS). A quantitative quasi-experimental design was applied across multiple intersections using a 12-month before–after window. Data were collected from video analytics (vehicle and pedestrian detection, tracking, violations, road conditions), adaptive signal logs, crash and injury records, near-miss indicators, and contextual variables such as weather and traffic volume. Analysis combined perception validation (mAP, tracking accuracy), time-series operational assessment, and Difference-in-Differences modeling to estimate safety impacts. Results show high perception reliability (mAP > 0.85) and significant operational improvements, including a 33% reduction in waiting time and 35% shorter queues. More importantly, red-light violations decreased by 39%, near-miss events by 45%, crash frequency by 42%, and severity index by 37%. The findings indicate a causal pathway from vision-based perception to adaptive control and enforcement, leading to measurable safety gains. The study concludes that computer vision serves as a safety governance instrument within Smart City ITS when detection outputs are tightly coupled with intervention mechanisms.