31st Florida Conference on Recent Advances in Robotics May 10-11, 2018, University of Central Florida, Orlando, Florida A Sensor for Visibility Determination under Fog Conditions Christina Drake, Harish Chintakunta, Christopher Coughlin, Ezequiel Garcia, Scott Hoos, Kristyn Ardrey, Paul Luckey and Aubury Erickson Florida Polytechnic University 4700 Research Way Lakeland, Florida 33805 cdrake@floridapoly.edu, hchintakunta@floridapoly.edu, ccoughlin@floridapoly.edu ABSTRACT In this paper, we describe a sensor approach utilizing commercial highway cameras for visibility determination under fog conditions. We achieved this by creating an engineered object in the field of view (FOV) of the camera that utilized two measurement approaches (contrast and light evaluation) via image processing. We employ the Koschmieder’s law to estimate visibility conditions from fog measurements, and scattering profile to estimate visibility conditions from visible measurements of a modulated light. The contrast measurements are used during the daytime, and light evaluations during night. We discuss the merits and challenges of such an approach. Keywords Weather Sensor, Image Processing 1. INTRODUCTION The onset of sudden fog or extremely dense fog is an issue that is critical to driver safety. Current State of the Art in real time fog detection is plagued by several false positive and false negative events due to the nature of fog formation, specifically when looking at low fog formation that would impact drivers. Both the thermodynamics and kinetics of fog formation make it difficult to evaluate over a small interrogation volume. Existing laser based sensors, such as the Vaisala PWD12, while excellent at close range measurement, are expensive, and are not able to get around the elusiveness of the type of fog that decreases driver visibility and detect with high confidence. Highway and other commercial cameras have the advantage of viewing a scene near a roadway and other areas of interest and typically have a large viewing area. We sought to take advantage of this by creating an object in the field of view of the camera and used it in conjunction with image processing for the purpose of visibility determination. There are similar studies that utilize image processing on commercial cameras for either visibility or fog determination. The contrast of a “black” target is one established way of making visibility determinations that has been accepted and widely used since the 1920s. Hautiere et al. [1] demonstrated an onboard vehicle camera for fog detection. Similar to our approach, this group utilized a variation of contrast evaluation in the FOV based on the treatment by Koschmeider [2]. This technique provided only an estimation of visibility, and was inoperable at night. Tang et al [3] also sought to create a contrast sensor utilizing a camera. They compared a black target to the horizon. At reduced visibilities, their approach began to deviate and see decreased correlation. This is likely because they used the horizon as the contrast media to the black and was not controlled in the measurement. In our approach, we also utilize Koschmieder’s law as a basis of our contrast measurement but utilized against a tightly engineered contrast piece. We also utilized an LED light (650 nm) to monitor scattering with increased fog density. 2. SENSOR APPROACH For the contrast portion of our initial sensor prototype, we utilized a high albedo, diffuse white surface with a near perfect black contrasting agent for visibility determination at varying fog densities. The departure from convention that makes our sensor unique is that a high albedo material (the front contrast pattern) is used in place of the sky or horizon as a proxy to ambient light or scene brightness, Figure 1. Figure 1. a) CAD of first sensor prototype and b) Sensor face and main components.