Development of a Road Visibility Forecast Model on the Basis of a CCTV-Camera-Based Road Visibility Information System

This study proposes a model that forecasts visibility in winter on the basis of a CCTV-Camera-Based Road Visibility Information System (RVIS) that was developed by the authors to minimize winter hazards, including traffic accidents resulting from snowstorm-induced poor visibility. The RVIS quantifies road images recorded by multiple closed-circuit television (CCTV) cameras along the roads to automatically apply the quantified information to pre-defined visibility indexes that indicate levels of road visibility. To quantify the road images, the weighed intensity of power spectrum (WIPS) is calculated such that its magnitude represents the difference in spatial frequencies. When road visibility decreases from snowstorm or fog, the contrast of road view contained in the CCTV camera image decreases. From WIPS values, the road visibility index (RVI) is calculated. The RVI is categorized into 4 ranks for greater understandability by users: Level 1(≧500 m), Level 2 (200~500 m), Level 3 (100~200 m) and Level 4(<100 m). From the RVI ranks, road users can know the visibility level on the route where they are planning to drive. The objective of this study is to develop a visibility forecast model utilizing the RVIS. To achieve the goal, we used the data of the WIPS values and RVI ranks recorded by the RVIS, as well as 1-km-mesh meteorological data recorded by the Japan Weather Association. A feasibility study on the visibility forecast model was conducted during the winter of 2009-2010 at a 35-km section of National Route 40 in Hokkaido, Japan. In the study, 1-km-mesh meteorological data were used to estimate the WIPS values and RVI ranks a few hours ahead of time by employing a multiple-regression model and the Kalman filter. The relationship between WIPS data from road images as dependent variables and the meteorological data as independent variables was examined to compare the accuracy of WIPS estimation between the two models. As a result, the correlation coefficient for the Kalman filter indicated better fit than that for the multiple-regression model. Thereby, the Kalman filter was identified to be more applicable than the multiple-regression approach to create a road visibility forecast model.