A Low-Complexity Vision-Based System for Real-Time Traffic Monitoring

Abstract

In this paper a novel, efficient, and fast-performing vision-based system for traffic flow monitoring is presented. Using standard traffic surveillance cameras and effectively applying simple techniques, the proposed method can produce accurate results on vehicle counting in different challenging situations, such as low-resolution videos, rainy scenes, and situations of stop-and-go traffic. Due to the simplicity of the proposed algorithm, the system is able to manage multiple video streams simultaneously in real time. The method follows a robust adaptive background segmentation strategy based on the Approximated Median Filter technique, which detects pixels corresponding to moving objects. Experimental results show that the proposed method can achieve sufficient accuracy and reliability while showing high performance rates, outperforming other state-of-the-art methods. Tests have proved that the system is able to work with up to 50 standard-resolution cameras at the same time in a standard computer, producing satisfactory results.