Real-Time Traffic Surveillance and Vehicle Speed Detection Using Machine Vision

Abstract

Traffic monitoring and the assessment of vehicle speeds are essential for effectively managing congestion, improving urban mobility, and enhancing road safety. Conventional traffic surveillance methods typically depend on intrusive sensors, such as inductive loops and radar systems, which are expensive, require significant installation efforts, and pose maintenance challenges, rendering large-scale implementation impractical. In this study, we introduce a non-invasive, real-time traffic monitoring system that utilizes computer vision techniques to efficiently detect, track, and estimate vehicle speeds. This system incorporates deep learning-based object detection and centroid tracking to accurately identify and monitor vehicles across multiple frames. A mapping function is employed to translate pixel-based measurements into real-world distances, ensuring precise speed calculations. Furthermore, the adaptive tracking algorithm improves the system's robustness against environmental changes, occlusions, and varying traffic conditions. Traffic density is assessed based on real-time variations in speed and the number of vehicles detected, offering comprehensive insights into traffic conditions. The proposed system was tested using actual traffic footage, revealing high accuracy in vehicle detection, tracking, and speed estimation. This framework is scalable, cost-effective, and applicable to a range of uses, including automated traffic control, law enforcement, and smart city projects. By removing the necessity for physical sensors and leveraging deep learning-based vision techniques, our approach represents a significant advancement in intelligent transportation systems, providing an efficient and sustainable solution for contemporary urban traffic management.