Wavelet Transform for Analyzing Fog Visibility

A GROWING NUMBER OF COMputer-vision systems have emerged for traffic analysis and control. Increased traffic, along with steady efforts to improve driver security, have encouraged the development of sophisticated video-based systems, providing statistical measures of traffic flow for actual traffic information on variable message signs. Investigations show that VMS reduce traffic jams, thus minimizing the burden placed on financial and environmental resources. Moreover, traffic control systems reduce the number of accidents to 45%.1 Of particular interest to drivers is information on low visibility in fog—a dangerous driving condition that causes many accidents. Because places where fog frequently rolls in are often well known, message signs inform motorists of visibility distance. Traditionally, the visibility distance is estimated with transmissiometers, but we suggest replacing conventional fog-detection sensors with the integrated functionality that a computer-vision system can provide. Our method, based on a B-spline wavelet transform, analyzes video frames used in traffic-control systems and derives a visibilitydistance measure with extended signal-analysis techniques. With our wavelet-based approach, we can inform drivers in real time of the presence of fog through VMS located a few kilometers before the foggy area. Analyzing the entire traffic scene globally, our method provides more precise input to fog information systems than the traditional transmissiometers that analyze fog density locally beside the roadway.