Real-Time Traffic Monitoring Using Low-Resolution Web Cameras
Musharaf Ali, Zuev Sergey, Anko Boerner · elib (German Aerospace Center) · 2012
This paper presents a novel approach for vision- based real-time traffic parameters estimation for a local section of road network from free, available, low-cost web cameras. Special investigation has been done for robustness and reliability of traffic parameters estimation algorithms. Processing chain includes: perspective correction of images, vehicle detection, velocity estimation, and vehicle counting. Perspective transformation removes perspective distortion, while lane markers on the road are used for scale factor calculation. For vehicle detection from traffic scenes, the authors developed a novel background estimation algorithm based on the median of First-in-First-out (FIFO) buffer filled with images using Mutual information, which is a self-adaptive algorithm according to the changing weather, illumination and traffic conditions. By making some assumptions and exploiting the domain knowledge of real-world traffic flow patterns, the vehicle counting and the velocity estimation can be determined reliably and accurately. The authors evaluate our approach for real- time traffic monitoring based on traffic web cameras available from Cologne city Germany. Results are highly encouraging and show promising potential for real-time traffic monitoring from low-resolution web cameras.