A Visibility Estimation Method Based on Digital Total-sky Images

The proposed visibility estimation method is a curve fitting algorithm, which establishes a relation between the image's atmospheric transmittance and atmospheric visibility. Firstly, the total-sky image is captured by the digital total-sky image visibility experimental platform. The core unit of this platform is a digital camera equipped with a fisheye lens, and the camera is placed vertically towards the sky. The platform can collect a total-sky image at specified time interval, and then the original image is transferred to a computer for image processing. In particular, the total-sky image needs to be converted into a panoramic image using an image calibration algorithm, and the panoramic image contains most of the near-surface image information. Next, the panoramic image is used to compute the visibility. Some visibility-related image features are extracted from the panoramic image firstly. The image's atmospheric transmittance can be calculated using dark channel prior theory. The relationship between the atmospheric transmittance and atmospheric visibility can be established by curve fitting method, and the initial visibility estimate model based on total-sky images is achieved. The model can be improved by combining a number of field experiments. Finally, the retrieved visibility is calculated by importing the real-time total-sky image into the model.Results show that the basic trend of visibility data from total-sky visibility estimation model is consistent with that of the forward scattering visibility meter through the comparative test and calculating correlation coefficients. The trend is most noticeable in low or medium visibility. However, as the visibility increases, the consistency decreases because of more fluctuation. As the forward scattering visibility meter used to establish the model whose measuring range is from 0 to 35 km, estimate model results are generally less than the measurement of forward scattering visibility meter especially when the visibility is high. In general, the basic trend of visibility data of total-sky visibility estimation model is consistent with that of the forward scattering visibility meter when the global atmospheric light is well-distributed and there is no underexposure or overexpose. The correlation coefficient between results of two methods is close to 1, which also means that the consistency between the two methods is good. In addition, the image features used in this method do not depend on a certain point in the image, nor are they limited to a certain range of visual distance. At the same time, there is no need to use manually set target or to fix a particular building, which makes it easier for observers to select the appropriate direction to measure visibility accurately. The proposed method has advantages of high measurement accuracy and large sampling range and can be used as a supplementary observation method of the traditional forward scattering visibility meter.