AUTOMATIC VISIBILITY MEASUREMENT METHODS BASED ON VIDEO CAMERAS
Taejun Kwon, E Fleege · Xth PIARC International Winter Road CongressPermanent International Association of Road Congresses · 1998
This paper describes several automatic visibility measurement methods which were cooperatively developed by the University of Minnesota, Duluth and the Minnesota Department of Transportation based on video image processing technologies. The first algorithm developed for daytime visibility is devised using the concept of dual targets that have identical contrast characteristics. This uses the principle that the degree of contrast reduction from one target to another follows the Koschmieder rule. At night, two targets with identical intensity of light sources are used to measure the degree of light intensity follows reduction from one target to the other. Since the attenuation of light intensity follows the Allard's law, visual range is directly derived from the relationship between the extinction coefficient of air and the degree of light intensity reduction registered to the image. Although, the measurement methods based on contrast work in principle, it produces a relatively large amount of error under non-uniform distribution of atmospheric attenuation coefficients in the target region. Hence, we developed a new algorithm that utilizes a statistical modeling analysis along the line of sight where localized edge information exponentially decreases as the distance increases. In order to evaluate the performance of the algorithms developed, a set of real-word highway data was used as the basic test bed. The experimental results include performance comparison of several commercial visual range instruments with the present video-image-based algorithms.